Uluslararası İleri Doğa International Journal of Advanced

Bilimleri ve Mühendislik Natural Sciences and Engineering

Araştırmaları Dergisi Researches
Sayı 9, S. 546-558, 12, 2025 Volume 9, pp. 546-558, 12, 2025
© Telif hakkı IJANSER’e aittir Copyright © 2025 IJANSER
Araştırma Makalesi [Link]
Research Article
ISSN:2980-0811

Process Optimization with Response Surface Methodology: A

Comprehensive Approach by Experimental Design Types
Khanda Khorsheed, Fatemeh Asadi, Roghaiyeh Asadi, Ercan Aydoğmuş,
Buket Erzen, Mukaddes Karataş, Şermin Deniz, and Ramazan Orhan

Department of Chemical Engineering, Faculty of Engineering, Fırat University, 23119, Elazığ, Türkiye
*(ercanaydogmus@[Link]) Email of the corresponding

(Received: 19 Deceember 2025, Accepted: 22 December 2025)

ATIF/REFERENCE: Khorsheed, K., Asadi, F., Asadi, R., Aydoğmuş, E., Erzen, B., Karataş, M., Deniz, Ş. & Orhan, R. (2025).
Process Optimization with Response Surface Methodology: A Comprehensive Approach by Experimental Design Types,
International Journal of Advanced Natural Sciences and Engineering Researches, 9(12), 546-558.

Abstract – Response surface methodology (RSM) is an advanced statistical optimization method developed
for analyzing multivariate systems and determining optimum process conditions within these systems. It is
widely used, particularly in fields such as chemical engineering, mechanical engineering, biotechnology,
food engineering, and quality control. RSM enables the prediction and improvement of system performance
by modeling the relationships between a dependent variable (response) and one or more independent
variables (factors). The foundation of this method lies in quadratic polynomial models created using
experimental data. These models yield more accurate results by accounting for interactions and non-linear
relationships between the factors. The design of experiments (DoE) plays a crucial role in RSM
applications. The primary experimental design types used for this purpose include central composite design
(CCD), Box-Behnken design (BBD), and full factorial designs. These designs allow the factors to be tested
at various levels, increasing the reliability of the model and enhancing its predictive capability. CCD is one
of the most commonly used design types in RSM. It generally consists of two main components: a factorial
(or fractional factorial) part and axial (star) points. CCD is specifically structured to effectively include
quadratic terms that capture the curvature of the response surface. Moreover, the inclusion of central points
allows for the assessment of experimental repeatability and the estimation of experimental error. CCD can
be combined with both full and fractional factorial designs, offering a flexible framework. BBD improves
the stability of model estimates by providing more data at intermediate levels of the factors. It is particularly
preferred in situations where experimental costs are high, as it requires fewer runs. On the other hand, full
factorial designs offer the highest level of accuracy by evaluating all possible combinations of factors.
However, due to the large number of experiments required, their implementation can be challenging.
Therefore, more efficient designs such as CCD or BBD are generally favored in RSM projects. One of the
main advantages of RSM is its ability to save both time and resources by allowing systematic
experimentation. In this respect, it is significantly more effective and reliable than traditional ‘one-variable-
at-a-time’ approaches. Furthermore, response surfaces can be visualized using the developed models,
providing decision-makers with an intuitive understanding of system behavior. Optimum points can be
visually identified through graphical analysis and contour plots. With its systematic, mathematical, and
visual approach, RSM is a highly versatile optimization tool.
Keywords – Response Surface Methodology, Central Composite, Box-Behnken, Full Factorial Design, Optimization.

546
 International Journal of Advanced Natural Sciences and Engineering Researches

I. INTRODUCTION
RSM is an advanced statistical optimization method specifically developed for the rigorous analysis of
complex multivariate systems and the precise determination of the optimum process conditions within these
operational frameworks. This methodology is fundamentally concerned with exploring the relationship
between various input variables and the resulting output performance, often referred to as the response. It
represents a systematic evolution from simple experimental approaches to a comprehensive,
mathematically grounded strategy for process improvement. The primary goal is to map the response
surface and navigate the factor space efficiently to find the conditions—whether maximizing, minimizing,
or targeting a specific value—that deliver the best outcome. This sophisticated approach ensures that
researchers move beyond mere guesswork to informed, data-driven decision-making, significantly
enhancing the reliability and efficiency of experimental work [1-5].
The genesis of RSM can be traced back to the pioneering work of George E. P. Box and K. B. Wilson in
the early 1950s, marking a significant milestone in statistical quality and process control. They recognized
the limitations of traditional factorial designs, which are better suited for screening effects than for
continuous optimization. The initial conceptualization involved a sequential approach: first, fitting a
simpler first-order model to quickly move toward the optimum region using the method of steepest ascent,
and subsequently, fitting a more complex second-order model once the neighborhood of the optimum was
reached. This sequential strategy underscores the methodology's efficiency, ensuring that costly and
complex experiments are only performed in the most promising area of the factor space. The historical
development solidified RSM's role as the premier technique for empirical modeling and optimization in the
scientific community [6-8].
The fundamental objective of utilizing RSM is to model and enhance system performance by establishing
clear, empirical relationships between the input independent variables (factors) and the dependent variable
(response). The system's performance, which is the response, is assumed to be a continuous function of the
factors under study. Therefore, the ultimate aim is not just to find the best point, but to understand the shape
of the entire response surface surrounding that optimal point. This understanding is critical for robust
process design, as it allows engineers to predict how the system will perform even under slight deviations
from the perfect optimal settings. This holistic view of the system’s behavior distinguishes RSM from
simple factor-level testing, providing a powerful predictive tool for industrial and academic research [9-
11].
RSM has achieved widespread acceptance and application, particularly within technical and scientific
disciplines that deal with complex chemical and physical transformations. It is a cornerstone of research in
chemical engineering, where it is instrumental in optimizing reaction yields, conversion rates, and the
selectivity of processes by adjusting parameters like temperature, pressure, and reactant concentrations.
The ability to model these multi-dimensional effects simultaneously allows for the fine-tuning of industrial
reactors, leading to improved product quality and reduced manufacturing costs. Furthermore, RSM is
routinely employed for the optimization of catalytic processes, determining the ideal catalyst formulation
or regeneration conditions, which is crucial for sustainable industrial operations [12-14].
Beyond the chemical sector, the methodology proves invaluable in fields like biotechnology, food
engineering, and quality control, showcasing its remarkable versatility. In biotechnology, RSM is
indispensable for optimizing fermentation processes, enhancing microbial growth, and maximizing the
yield of desired biomolecules, such as enzymes or antibiotics, by adjusting parameters like pH, nutrient
concentration, and agitation speed. Food engineering relies on RSM to optimize crucial quality attributes
like texture, color, and shelf life, ensuring consumer satisfaction and minimizing waste. Similarly, in quality
control and analytical chemistry, RSM is used to optimize measurement procedures, such as extraction or
chromatographic separation, thereby improving the accuracy and precision of analytical results [15-17].
The foundation of the RSM approach lies in the quadratic polynomial models constructed using the
carefully gathered experimental data. In the optimization phase, a second-order model is typically employed
to accurately capture any existing curvature in the response surface, which is indicative of complex factor

547
 International Journal of Advanced Natural Sciences and Engineering Researches

interactions and non-linear effects. The statistical significance of these coefficients is meticulously tested
through analysis of variance (ANOVA), ensuring that only meaningful terms remain in the final predictive
equation [18-20].
These sophisticated models yield more accurate and reliable results precisely because they systematically
account for the often-overlooked but crucial interactions and non-linear relationships among the factors. A
simple linear model, while easy to construct, would fail dramatically if the effect of Factor A changes
significantly depending on the level of Factor B. The inclusion of the cross-product term is what allows the
model to capture this synergistic or antagonistic behavior, providing a realistic depiction of the process
mechanics. Consequently, the quadratic model provides a much better fit to the observed data and offers
superior predictive capability within the specified experimental region [21-23].
The core principle behind this enhanced modeling capacity is the shift from a local linear approximation to
a more global quadratic description of the response surface. When the experimental region is near the
optimum, the response surface is characterized by a ridge or a stationary point (a maximum, minimum, or
saddle point). Only a model containing quadratic terms can correctly model this curvature and accurately
locate the stationary point, which represents the optimal process condition. Without the second-order terms,
a researcher might mistakenly conclude that the optimum lies beyond the tested region, leading to erroneous
or suboptimal process settings [24,25].
A fundamental part of successfully applying RSM is the proper utilization of design of experiments (DoE)
principles. DoE serves as the critical blueprint for organizing the physical experiments, ensuring that the
collected data is statistically sound and maximized for informational value. The chosen design dictates the
number of experiments required, the specific combinations of factor levels to be tested, and the ability of
the resulting data to resolve the coefficients of the polynomial model. A poorly chosen design can lead to
an ill-fitting model with high prediction variance, completely negating the benefits of the RSM approach
[26,27].

II. METHODOLOGY
Basic Concepts of RSM
RSM is built upon several interrelated foundational elements that facilitate process modeling, analysis, and
optimization. Understanding these concepts is essential for effectively applying RSM in experimental
research. Figure 1 shows the approach and models proposed in experimental design in RSM.

Figure 1. Experimental designs and models proposed in the RSM basic concept
548
 International Journal of Advanced Natural Sciences and Engineering Researches

The strategic application of DoE in RSM is paramount, as the design's structure directly influences the
reliability of the model and its subsequent predictive capability. By systematically varying the factors at
pre-determined levels according to the chosen design, the experimenter ensures that all necessary
information to estimate the main, interaction, and quadratic effects is gathered. This structured approach
contrasts sharply with haphazard experimentation, where correlation and causation can easily be confused.
The selection among the primary experimental design types—Central Composite Design (CCD), Box-
Behnken Design (BBD), and Full Factorial Designs—is the most critical decision in the initial stage of any
RSM project [28].

Central Composite Design (CCD)

CCD stands out as one of the most widely used and versatile design types in RSM, particularly favored for
its high efficiency in fitting a second-order model. Its structure is ingeniously crafted to provide maximum
information with a manageable number of runs, making it an excellent compromise between the exhaustive
nature of a full factorial design and the limited scope of a first-order screening design. The design is
specifically constructed to be rotatable or near-rotatable, meaning the prediction variance is constant at all
points equidistant from the center of the design, which significantly improves the model's reliability across
the experimental domain [29].
The CCD is characteristically composed of three main components: a factorial (or fractional factorial) part,
axial (star) points, and center points. The factorial portion, often a 2k design, allows for the efficient
estimation of the main effects and the two-factor interaction terms, defining the cubic boundaries of the
design space. The strategic placement of the axial points, lying at a specified distance (α) along the axis of
each factor, is what introduces the necessary levels to compute the quadratic terms. This unique
configuration is essential for effectively capturing the curvature of the response surface, which is the
hallmark of any optimization study [30,31].
The inclusion of the axial (star) points is the differentiating feature that enables the CCD to handle the non-
linear second-order model accurately. These points are specifically situated to allow for the estimation of
the pure quadratic effect for each factor by measuring the response at the extreme high and low settings of
that factor while all other factors are held at their center level. The distance of these axial points from the
center, denoted by α, is often set to achieve the desirable property of rotatability, ensuring that the precision
of the prediction is uniform throughout the design space. Properly calibrated axial points are therefore
indispensable for deriving a robust and well-behaved quadratic model [32,33].
The design also incorporates multiple central points—experiments where all factors are set at their midpoint
level—which serve two vital statistical functions. Firstly, these replicated center runs are used to assess the
experimental repeatability or pure error of the measurement system, providing a baseline against which the
variation attributed to the model terms can be compared. Secondly, and equally important, the difference
between the mean response at the center points and the mean response at the factorial points is used to
calculate the coefficient of the quadratic terms, providing a direct indication of the curvature in the model
[34].
The flexibility of the CCD is further demonstrated by its ability to be constructed using either full factorial
designs or fractional factorial designs for its corner points. When the number of factors is large (e.g., k ≥
0), a fractional factorial design can be used to reduce the overall number of runs significantly while still
allowing for the robust estimation of all necessary main and second-order effects, provided that higher-
order interactions are negligible. This adaptability makes CCD an efficient choice, balancing the need for
precise optimization modeling with the practical constraints of experimental resources, maintaining a
framework that is both statistically powerful and logistically feasible [35].
However, the selection of the α value in CCD is a critical engineering decision, as setting it too large might
place the axial points outside the practical or safe operating range of the physical system. Therefore, careful
consideration must be given to the physical constraints of the process before the design is finalized. If the
desired α value, which ensures rotatability, results in physically unattainable factor combinations, a

549
 International Journal of Advanced Natural Sciences and Engineering Researches

compromise must be made, possibly sacrificing some degree of rotatability for a design that is physically
executable. This ensures the integrity of the experiment and the relevance of the derived optimal conditions
to real-world application [36].
Figure 2 illustrates the axial (star) points defined by the α (alpha) values in a CCD. In this representation,
the experimental domain is extended beyond the factorial region by positioning axial points symmetrically
at +α and −α along each factor axis, while keeping the remaining factors at their central levels. These axial
points are essential for estimating the curvature of the response surface and enabling the development of a
quadratic polynomial model within RSM. By expanding the design space, the α points improve the
prediction accuracy near the optimum and allow reliable identification of interaction and second-order
effects. The symmetric placement of the axial points ensures rotatability, meaning that the variance of the
predicted response remains constant at equal distances from the design center. This graph visually
emphasizes the role of α values in enhancing model robustness, sensitivity to curvature, and optimization
efficiency in CCD-based experimental designs [37,38].

Figure 2. Central Composite Design (CCD)

Box-Behnken Design (BBD)

BBD represents another highly regarded and frequently employed class of rotatable designs in RSM,
offering a compelling alternative to the CCD. BBD is particularly characterized by its experimental points
residing at the midpoints of the edges of the factor space cube and at the center, meaning the design avoids
all factor combinations at the corners (vertices) of the design region. This characteristic makes it highly
valuable in applications where extreme factor combinations, which correspond to the corners of the cube,
are either physically unachievable, potentially hazardous, or prohibitively expensive.
A key advantage of the BBD is its inherent efficiency in terms of the number of experimental runs required,
especially when dealing with four or more factors. This makes the BBD an economically sound choice and
is particularly preferred in situations where experimental costs are exceptionally high, such as in expensive
pilot plant studies or time-consuming biological assays. By providing a sufficient number of data points at
the intermediate levels of the factors and around the center, BBD significantly improves the stability of the
model estimates and reduces the likelihood of parameter correlation, leading to a more robust and reliable
predictive equation.
In contrast to the geometrically comprehensive but resource-intensive nature of CCD and BBD, full
factorial designs provide the ultimate in experimental accuracy by executing every single possible
combination of factor levels. For an RSM application requiring the testing of factors at three distinct levels
(low, medium, and high), a full factorial design necessitates 3k runs, where k is the number of factors. This
exhaustive evaluation ensures that every possible interaction and non-linear effect is unequivocally
captured and estimated without any ambiguity, providing the highest possible resolution of the response
550
 International Journal of Advanced Natural Sciences and Engineering Researches

surface.
The significant drawback of full factorial designs, however, lies in the large number of experiments
required, which grows exponentially as the number of independent variables increases. For example, a
system with only five factors requires 35 = 243 runs, and a system with six factors jumps to 36 = 729 runs.
This exponential increase makes the implementation of full factorial designs challenging and often
impractical from the perspectives of time, labor, material consumption, and overall cost.
Figure 3 represents the BBD used in RSM. The design points are located at the midpoints of the edges of
the experimental space and at the center point, while the corner points are deliberately excluded, ensuring
safe and practical experimental conditions. Color-coded markers illustrate the distribution of experimental
runs across the factor levels (Factor A, Factor B, Factor C), and smooth response surfaces or contour plots
highlight the quadratic relationship between factors and responses. The figure emphasizes that BBD
efficiently captures main effects, interaction effects, and curvature with a reduced number of experiments,
making it ideal for optimization studies in engineering and material science [39,40].

Figure 3. Box-Behnken Design (BBD)

The engineering and statistical community has generally favored more efficient designs such as CCD or
BBD in the vast majority of RSM projects. These designs are statistically tailored to provide all the
necessary information for fitting a second-order model—the main effects, two-way interactions, and
quadratic effects—while drastically minimizing the total experimental burden. This strategic compromise
between experimental size and model fidelity underscores the modern approach to process optimization,
prioritizing systematic and cost-effective data collection over exhaustive but unwieldy experimentation.
One of the most profound main advantages of RSM is its ability to save both time and resources by
compelling the researcher to adopt a systematic experimentation approach. Unlike the archaic and
statistically invalid ‘one-variable-at-a-time’ methodology, RSM's designs ensure that variables are changed
concurrently and in a coordinated fashion, maximizing the information yield per experiment. By
minimizing the total number of runs while still achieving a high-fidelity model, RSM accelerates the
discovery and optimization phases of any research or industrial project. This efficiency translates directly
into lower operating costs and a faster time-to-market for new or improved products [41].
As shown in Figure 4, the systematic design of the RSM, the path to achieving optimum results, and the
OFAT approach are illustrated. The fundamental flaw of OFAT is its inability to detect interaction effects;
if the effect of Factor A depends on the level of Factor B, OFAT will only report misleading, simple-factor
main effects. RSM, by contrast, is specifically designed to simultaneously estimate all factor main effects
and their critical interaction terms, providing a statistically sound and scientifically accurate map of the
factor space.
The immense utility of RSM is amplified by its capability to produce visual representations of the response
surfaces using the developed polynomial models. These contour plots and 3D surface plots provide
decision-makers with an intuitive understanding of system behavior, making complex mathematical
relationships accessible to non-statistical personnel. The optimum points can be visually identified through
the peaks or troughs on these graphical analyses, allowing for quick verification and communication of the
results. This seamless integration of a systematic, mathematical, and visual approach confirms RSM's status
551
 International Journal of Advanced Natural Sciences and Engineering Researches

as a highly versatile and essential optimization tool in the contemporary engineering and scientific
landscape.

Figure 4. The systematic design of RSM, the path to achieving optimum, and the OFAT approach.

Full Factorial Designs (FFD) in RSM

Full factorial designs constitute one of the most rigorous and information-rich experimental strategies
employed in RSM. In these designs, all possible combinations of factor levels are systematically evaluated,
providing complete coverage of the experimental domain. This exhaustive structure enables researchers to
identify not only the individual main effects of each factor but also the full range of first-order and higher-
order interaction effects, which are essential for accurately describing the behavior of complex
multivariable systems. When factors are studied at three levels—commonly coded as −1, 0, and +1—the
design takes the form of a 3ᵏ factorial, allowing the experimenter to capture curvature, nonlinear behavior,
and synergistic or antagonistic interactions with unmatched precision. As a result, full factorial designs
serve as a methodological benchmark for accuracy and model fidelity in optimization studies.
Despite their superior resolution, full factorial designs can quickly become resource-intensive due to the
exponential growth in the number of required runs as the number of factors increases. For instance, a system
with four factors requires 3⁴ = 81 experimental runs, while one with six factors demands 3⁶ = 729 runs. This
rapid escalation imposes significant demands on time, labor, materials, and operational costs, which may
render the approach impractical for many experimental scenarios, especially those involving high-cost
materials, lengthy processing steps, or limited access to specialized equipment. These constraints have
motivated the development and adoption of more efficient designs—such as the CCD and the BBD—which
reduce the total number of trials while still capturing the essential model properties required for second-
order optimization.
FFD remain an invaluable tool when precision is paramount and sufficient resources are available. Their
ability to provide an unambiguous estimation of all main and interaction effects makes them particularly
advantageous in early-stage mechanistic studies, process characterization, and systems where factor
relationships are suspected to be highly nonlinear or strongly coupled. Furthermore, full factorial datasets
form a robust foundation for constructing high-quality predictive models, performing sensitivity analyses,
validating simplified experimental designs, and benchmarking alternative optimization approaches. In
cases where experimental resources permit, the comprehensive nature of full factorial designs ensures
maximum statistical power and delivers the most reliable insight into the underlying dynamics of
multivariate engineering and scientific systems [42].
Figure 5 illustrates the concept of FFD used in experimental design and RSM. In a full factorial design, all
possible combinations of factor levels are systematically tested, resulting in a complete and structured grid
of experimental points across the design space. Each axis represents an experimental factor (Factor A,
Factor B, Factor C), while the colored points indicate experiments conducted at low, medium, and high
552
 International Journal of Advanced Natural Sciences and Engineering Researches

levels. This comprehensive coverage allows precise evaluation of main effects, interaction effects, and
combined factor influences without ambiguity. The visual emphasizes that, although full factorial designs
require a larger number of experiments, they provide the most detailed and reliable understanding of factor–
response relationships, making them especially valuable in fundamental studies, mechanism analysis, and
early-stage process development.

Figure 5. Full Factorial Designs (FFD)

III. RESULTS AND DISCUSSIONS

The synthesized literature and methodological analysis clearly demonstrate that RSM offers a substantial
improvement over traditional one-factor-at-a-time experimentation in both efficiency and scientific rigor.
Across the surveyed applications in chemical engineering, biotechnology, food engineering, and quality
control, RSM consistently enabled researchers to identify optimal or near-optimal operating conditions with
markedly fewer experimental runs than classical approaches would require. The conceptual framework
summarized in Figures 1–5 highlights how the combination of statistically sound design of experiments
(DoE), quadratic modeling, and visual response surfaces transforms process optimization from a trial-and-
error activity into a structured, data-driven procedure. In particular, the ability of RSM to simultaneously
quantify main effects, interaction effects, and curvature provides a robust empirical description of complex
multivariate systems, supporting both mechanistic insight and practical process improvement.
A central outcome of this review is the comparative evaluation of the three principal design classes used in
RSM: CCD, BBD, and FFD. CCD emerges as the most versatile and frequently used second-order design,
owing to its flexible structure composed of factorial (or fractional factorial) points, axial (star) points, and
center points. This configuration ensures that the model parameters associated with linear, interaction, and
quadratic terms are all estimable with reasonable precision, while maintaining a manageable number of
runs for typical engineering systems with two to six factors. BBD, by contrast, strategically avoids the
extreme corners of the design space, making it especially attractive in scenarios where operating at
boundary conditions is unsafe, economically prohibitive, or technically infeasible. Full factorial designs at
three levels provide the most complete and unbiased information but become computationally and
experimentally prohibitive as the number of factors increases. The comparative analysis therefore confirms
that CCD and BBD offer an optimal compromise between information richness and experimental economy,
which explains their widespread dominance in modern RSM-based optimization studies [43].
The examination of model-building practices further underscores the superiority of second-order
polynomial models for describing real-world processes within a limited experimental region. In most of the
reviewed applications, first-order models served only as preliminary tools for locating an approximate
553
 International Journal of Advanced Natural Sciences and Engineering Researches

region of interest; accurate optimization required the inclusion of quadratic and interaction terms. ANOVA,
lack-of-fit tests, and diagnostic residual plots commonly indicated that models without quadratic
components were inadequate, especially where strong curvature or synergistic/antagonistic interactions
were present between factors such as temperature, concentration, and residence time. Second-order models,
by contrast, systematically reduced prediction error, improved coefficients of determination (R2 and
adjusted R2), and produced well-behaved response surfaces suitable for graphical interpretation. The
resulting contour and 3D surface plots, as illustrated conceptually in Figures 2–5, proved particularly useful
for identifying stationary points, visualizing trade-offs between factors, and communicating optimal
operating windows to non-statistical stakeholders.
Another important outcome pertains to the practical implementation issues that frequently arise in real
experimental environments. Although CCD offers theoretical rotatability when the axial distance α is
selected according to established formulas, physical or safety constraints often limit the feasible factor
ranges, forcing compromises in design geometry. The reviewed studies show that when α is reduced to
keep all experimental points within realistic operating boundaries, a slight loss in rotatability is generally
acceptable and does not negate the benefits of the design, provided that adequate replication at the center
points is maintained. These center runs play a crucial role in estimating pure error, detecting curvature, and
stabilizing the model. Furthermore, the results emphasize that careful coding of variables, appropriate
scaling, and rigorous pre-experimental screening of factors are essential to prevent aliasing, reduce model
uncertainty, and ensure that RSM is applied within a region where the quadratic approximation remains
valid.
The integrative analysis of recent literature indicates a clear trend toward hybrid strategies that couple RSM
with advanced computational intelligence techniques such as artificial neural networks, adaptive neuro-
fuzzy inference systems, and gradient-boosting models. In these frameworks, RSM often serves as a
systematic tool for initial design and factor screening, while machine learning models refine predictions in
high-dimensional or highly nonlinear domains. This synergy is particularly promising for applications in
non-conventional energy systems, complex biological processes, and climate-related modeling, where data
are expensive and system dynamics are intricate. At the same time, multi-response optimization using
desirability functions remains an active area of development, enabling simultaneous optimization of
multiple performance criteria such as yield, selectivity, energy consumption, and product quality.
Collectively, these findings confirm that RSM—especially when combined with robust designs such as
CCD and BBD—has evolved into a central pillar of modern experimental design and process optimization,
and its relevance is likely to increase further as it is increasingly integrated with emerging data-driven and
AI-based methodologies [44].
Figure 6 shows the progressive methodological evolution from traditional OFAT experimentation toward
RSM, emphasizing how statistically designed experiments enhance predictive accuracy and process
understanding. The transition from scattered, unsystematic OFAT data to structured RSM modeling
highlights RSM’s ability to fit quadratic response surfaces that capture nonlinear interactions between
variables, thereby enabling precise optimization across multidimensional factor spaces. The comparative
design panel further distinguishes the capabilities of CCD, BBD, and FFD, showing how each provides
different levels of experimental resolution and curvature detection. Practical implementation is represented
by the identification of process optima and the integration of machine learning (ML), demonstrating how
modern analytical tools extend RSM’s predictive strength by enabling deeper pattern recognition and
generalization beyond the experimental domain. Finally, the hybrid RSM+ML strategy underscores a
contemporary trend in materials science and engineering research, where statistically efficient design of
experiments is coupled with data-driven computational intelligence to achieve robust, high-accuracy
optimization and mechanistic insight simultaneously.

554
 International Journal of Advanced Natural Sciences and Engineering Researches

Figure 6. Methodological evolution, statistical evaluation, and machine learning progressing from traditional OFAT
experiments to RSM.
The comparison between the OFAT and RSM clearly illustrates the superiority of multivariate
experimentation, where simultaneous factor variation captures interactions and nonlinear trends that OFAT
inherently overlooks. The upward, smooth response trend associated with RSM emphasizes how quadratic
modeling improves prediction accuracy and optimization reliability. This visual comparison reinforces the
discussion that RSM provides a structured, statistically valid pathway to reach optimal conditions with
fewer experiments and higher information density, which is a critical advantage in resource-intensive
engineering and scientific studies.
The geometric representations highlight how CCD combines corner points, axial points, and center points
to effectively estimate second-order models, while BBD strategically avoids extreme factor combinations,
enhancing safety and cost efficiency. The full factorial schematic, though information-rich, visually
conveys the rapid growth in experimental complexity as factor levels increase. This side-by-side depiction
supports the discussion section’s conclusion that CCD and BBD offer an optimal balance between
experimental economy and model fidelity, whereas full factorial designs, despite their accuracy, are often
impractical for large multivariable systems.
The lower portion of the figure emphasizes the conceptual foundation and future direction of RSM through
the visualization of quadratic response surfaces and hybrid optimization strategies. The three-dimensional
surface and contour representation demonstrates how optimal operating regions and stationary points can
be intuitively identified, supporting data-driven decision-making beyond purely numerical outputs.
Additionally, the integration of RSM with machine learning (RSM + ML) reflects a growing trend toward
hybrid modeling frameworks, where RSM provides systematic experimental design and interpretability,
while machine learning enhances predictive capability in complex or highly nonlinear systems. Overall, the
graphical abstract successfully encapsulates the methodological strengths, practical implications, and
evolving role of RSM, reinforcing its position as a cornerstone of modern process optimization research
[45].

555
 International Journal of Advanced Natural Sciences and Engineering Researches

IV. CONCLUSIONS
This comprehensive review underscores the critical role of RSM as a paramount statistical tool for the
optimization of complex, multivariate systems across various scientific and engineering disciplines. By
moving beyond the limitations of univariate experimentation, RSM provides a rigorous and resource-
efficient framework for modeling and understanding the intricate relationships, including non-linear effects
and factor interactions, that govern process performance. The fundamental strength of RSM lies in its ability
to approximate the true response function using empirical second-order polynomial models, which allows
researchers and engineers to accurately predict system behavior within a defined experimental region. The
systematic nature of RSM ensures that optimal operating conditions are identified reliably and efficiently,
leading to enhanced product quality, reduced operational costs, and accelerated process development.
The choice of the appropriate Design of Experiments (DoE) within the RSM framework is instrumental to
the success of the optimization effort. Among the available designs, the CCD stands out due to its optimal
balance between statistical power and experimental economy. Its ingenious structure, comprising factorial
points, strategically placed axial (star) points, and replicated center points, ensures that sufficient data is
collected to estimate all necessary first-order, interaction, and crucial quadratic coefficients. This structure
is specifically tailored to accurately capture the curvature of the response surface, which is essential for
locating the true stationary point (maximum, minimum, or saddle point). Furthermore, the rotatability and
flexibility of the CCD, particularly its ability to incorporate fractional factorial cores, cement its position
as the preferred design for most second-order optimization studies in chemical, mechanical, and
biotechnological research.
RSM is a valuable statistical tool used to optimize processes and systems by modeling the relationship
between multiple input variables and their effect on a response variable. Through the design of experiments,
RSM identifies the most significant factors influencing a system's output, enabling efficient optimization
with fewer experimental trials. RSM is particularly beneficial for complex systems where multiple factors
interact, and its ability to model these relationships in polynomial form provides a clear understanding of
how changes in inputs impact the outcome. The method is widely used across various industries, including
engineering, manufacturing, and healthcare, where improving efficiency and performance is crucial. While
RSM offers several advantages, such as reducing experimentation costs and time, it does require careful
experimental design and a solid understanding of the system under study. Its application may also be limited
when dealing with highly nonlinear or highly complex systems. In conclusion, RSM is an essential tool for
process optimization, allowing researchers and engineers to make data-driven decisions, reduce costs, and
enhance system performance. Its effectiveness in exploring multivariable interactions and optimizing
system responses makes it a cornerstone of modern experimental design.
In summary, RSM, supported by robust designs like the CCD, represents the state-of-the-art approach for
empirical process modeling. The methodology not only yields a predictive mathematical equation but also
provides an intuitive, visual understanding of the process landscape through contour and 3D surface plots,
aiding in critical decision-making. Future research efforts should continue to focus on applying these
techniques to increasingly complex systems, such as non-conventional energy processes or highly dynamic
biological systems, and on integrating them with advanced machine learning techniques to further enhance
their predictive accuracy and applicability in high-dimensional optimization problems.

ACKNOWLEDGEMENTS
This study was conducted at the Memnune Bildik Laboratory in Chemical Engineering Department at
Fırat University.

556
 International Journal of Advanced Natural Sciences and Engineering Researches

REFERENCES
1. Bezerra, M.A., Santelli, R.E., Oliveira, E.P., Villar, L.S. and Escaleira, L.A. Response surface methodology (RSM) as a
tool for optimization in analytical chemistry. 2008. Talanta, 76(5): p. 965–977.
2. Reji, M. and Kumar, R. Response surface methodology (RSM): An overview to analyze multivariate data. 2022. Indian
Journal of Microbiology Research, 9: p. 241–248.
3. Lamidi, S., Olaleye, N., Bankole, Y., Obalola, A., Aribike, E. and Adigun, I. Applications of response surface methodology
(RSM) in product design, development, and process optimization. 2022. IntechOpen: p. 1–28.
4. Myers, R.H., Montgomery, D.C. and Anderson-Cook, C.M. Response surface methodology: Process and product
optimization using designed experiments. 2016. John Wiley & Sons, Hoboken, USA.
5. Silva, V. and Cardoso, J.S. (eds.) Response Surface Methods – Theory, Applications and Optimization Techniques. 2024.
BoD – Books on Demand, Norderstedt, Germany.
6. Asoo, H.R., Alakali, J.S., Ikya, J.K. and Yusufu, M.I. Historical background of response surface methodology. 2024. In:
Response Surface Methods – Theory, Applications and Optimization Techniques. IntechOpen: p. 1–16.
7. Hill, W.J. George Box: His interface with industry and its impact. 2014. Applied Stochastic Models in Business and
Industry, 30(1): p. 11–19.
8. Anderson, M.J. and Whitcomb, P.J. RSM simplified: Optimizing processes using response surface methods for design of
experiments. 2016. Productivity Press, New York, USA.
9. Elfghi, F.M. A hybrid statistical approach for modeling and optimization of RON using response surface methodology and
artificial neural network. 2016. Chemical Engineering Research and Design, 113: p. 264–272.
10. Said, K.A.M. and Amin, M.A.M. Overview on the response surface methodology in extraction processes. 2015. Journal of
Applied Science & Process Engineering, 2(1): p. 1–12.
11. Mostafaei, M., Javadikia, H. and Naderloo, L. Modeling the effects of ultrasound power and reactor dimension on the
biodiesel production yield: Comparison of prediction abilities between response surface methodology and adaptive neuro-
fuzzy inference system. 2016. Energy, 115: p. 626–636.
12. Bezerra, M.A., Santelli, R.E., Oliveira, E.P., Villar, L.S. and Escaleira, L.A. Response surface methodology as a tool for
optimization in analytical chemistry. 2008. Talanta, 76(5): p. 965–977.
13. Susaimanickam, A., Manickam, P. and Joseph, A.A. A comprehensive review on RSM-coupled optimization techniques and
its applications. 2023. Archives of Computational Methods in Engineering, 30(8): p. 4831–4853.
14. Kuipers, J.A.M. and van Swaaij, W.P.M. Application of computational fluid dynamics to chemical reaction engineering.
1997. Reviews in Chemical Engineering, 13(3): p. 1–118.
15. Liu, Y., Dar, B.N., Makroo, H.A., Aslam, R., Martí-Quijal, F.J., Castagnini, J.M. and Barba, F.J. Optimizing recovery of
high-added-value compounds from complex food matrices using multivariate methods. 2024. Antioxidants, 13(12): p. 1510.
16. Malekjani, N. and Jafari, S.M. Food process modeling and optimization by response surface methodology. 2020. In:
Mathematical and Statistical Applications in Food Engineering. CRC Press: p. 181–203.
17. Istiqomah, A., Saputra, O.A., Firdaus, M. and Kusumaningsih, T. Response surface methodology as an excellent tool for
optimizing sustainable food packaging: A review. 2024. Journal of Biosystems Engineering, 49(4): p. 434–452.
18. Lenth, R.V. Response-surface methods in R, using rsm. 2010. Journal of Statistical Software, 32: p. 1–17.
19. Wei, J. A simplified model of quadratic polynomials in response surface methodology and its design of experiment. 2018.
In: Proceedings of the 3rd International Conference on Modelling, Simulation and Applied Mathematics (MSAM 2018).
Atlantis Press: p. 11–15.
20. Peng, Y., Khaled, U., Al-Rashed, A.A., Meer, R., Goodarzi, M. and Sarafraz, M.M. Potential application of response
surface methodology for prediction and optimization of thermal conductivity of aqueous CuO nanofluid. 2020. Physica A:
Statistical Mechanics and its Applications, 554: p. 124353.
21. Sham, F.A.F., El-Shafie, A., Jaafar, W.Z.W., Sherif, M. and Ahmed, A.N. Advances in AI-based rainfall forecasting: Past,
present and future directions with intelligent data fusion and climate change models. 2025. Results in Engineering, 27: p.
105774.
22. Tahmasbi, V., Qasemi, M. and Rabiee, A.H. Predicting maximum process temperature in cortical bone milling using
XGBoost. 2025. Iranian Journal of Materials Forming, 12(3): p. 4–19.
23. Toms, D., Deardon, R. and Ungrin, M. Climbing the mountain: Experimental design for efficient optimization of stem cell
bioprocessing. 2017. Journal of Biological Engineering, 11(1): p. 35.
24. Bezerra, M.A., Santelli, R.E., Oliveira, E.P., Villar, L.S. and Escaleira, L.A. Response surface methodology as a tool for
optimization in analytical chemistry. 2008. Talanta, 76(5): p. 965–977.
25. McDonald, D.B., Grantham, W.J., Tabor, W.L. and Murphy, M.J. Global and local optimization using radial basis
function response surface models. 2007. Applied Mathematical Modelling, 31(10): p. 2095–2110.
26. Beg, S., Swain, S., Rahman, M., Hasnain, M.S. and Imam, S.S. Application of design of experiments in pharmaceutical
product and process optimization. 2019. In: Pharmaceutical Quality by Design. Academic Press: p. 43–64.
27. Bezerra, M.A., Santelli, R.E., Oliveira, E.P., Villar, L.S. and Escaleira, L.A. Response surface methodology as a tool for
optimization in analytical chemistry. 2008. Talanta, 76(5): p. 965–977.
28. Granato, D. and de Araújo Calado, V.M. The use and importance of design of experiments in process modelling in food
science and technology. 2014. Mathematical and Statistical Methods in Food Science and Technology, 1: p. 1–18.
29. Tamandani, M. and Hashemi, S.H. Central composite and Box–Behnken designs for optimization of a molecularly
imprinted polymer-based micro-solid phase extraction method. 2022. Analytical Letters, 55(15): p. 2394–2408.

557
 International Journal of Advanced Natural Sciences and Engineering Researches

30. Stamenković, O.S., Kostić, M.D., Radosavljević, D.B. and Veljković, V.B. Comparison of Box–Behnken, face central
composite and full factorial designs in optimization of hempseed oil extraction. 2018. Periodica Polytechnica Chemical
Engineering, 62(3): p. 359–367.
31. Aydoğmuş, E., Aydın, M. and Arslanoğlu, H. Production and characterization of microsphere reinforced polyester
composite: Modeling of thermal decomposition with ANN and optimization studies by RSM. 2024. Petroleum Science and
Technology, 42(12): 1505–1521.
32. Sharifi, S., Nabizadeh, R., Akbarpour, B., Azari, A., Ghaffari, H.R., Nazmara, S. and Yousefi, M. Modeling and optimizing
parameters affecting hexavalent chromium adsorption from aqueous solutions using Ti-XAD7 nanocomposite: RSM-CCD
approach, kinetic, and isotherm studies. 2019. Journal of Environmental Health Science and Engineering, 17(2): 873–888.
33. Dağ, M., Aydoğmuş, E., Yalcin, Z.G. and Arslanoğlu, H. Diatomite reinforced modified safflower oil-based epoxy
biocomposite production: Optimization with RSM and assessment of outcomes by ANN. 2023. Materials Today
Communications, 35: 106327.
34. Danbaba, N., Nkama, I. and Badau, M.H. Application of response surface methodology (RSM) and central composite
design (CCD) to optimize minerals composition of rice–cowpea composite blends during extrusion cooking. 2015.
International Journal of Food Science and Nutrition Engineering, 5(1): 40–52.
35. Bayuo, J., Abukari, M.A. and Pelig-Ba, K.B. Optimization using central composite design (CCD) of response surface
methodology (RSM) for biosorption of hexavalent chromium from aqueous media. 2020. Applied Water Science, 10(6): 1–
12.
36. Pilkington, J.L., Preston, C. and Gomes, R.L. Comparison of response surface methodology (RSM) and artificial neural
networks (ANN) towards efficient extraction of artemisinin from Artemisia annua. 2014. Industrial Crops and Products, 58:
15–24.
37. Aseibichin, C., Ulakpa, W.C., Omenogor, I., Doyah, E., Olaseinde, A.A., Anakpoha, O.C. and Karuppannan, S. Modeling
and optimization of transesterification of Jatropha oil to fatty acid methyl ester: Application of response surface
methodology (CCD) and Taguchi orthogonal method. 2024. RSC Advances, 14(17): 11784–11796.
38. Aydoğmuş, E., Demirpolat, A.B. and Arslanoğlu, H. Isothermal and non-isothermal drying behavior for grape (Vitis
vinifera) by new improved system: Exergy analysis, RSM, and modeling. 2022. Biomass Conversion and Biorefinery, 12(2):
527–536.
39. Maheswari, C., Ramya, A.S., Priya, B.M., Sudhahar, S., Prabhu Raj, B., Lokesh, B. and Ramani, G. Analysis and
optimization on the biodegradable plate making process parameters using RSM-based Box–Behnken design method. 2021.
Journal of Material Cycles and Waste Management, 23(6): 2255–2265.
40. Szpisják-Gulyás, N., Al-Tayawi, A.N., Horváth, Z.H., László, Z., Kertész, S. and Hodúr, C. Methods for experimental
design, central composite design and the Box–Behnken design to optimise operational parameters: A review. 2023. Acta
Alimentaria, 52(4): 521–537.
41. Madondo, N.I. and Chetty, M. Anaerobic co-digestion of sewage sludge and bio-based glycerol: Optimisation of process
variables using one-factor-at-a-time (OFAT) and Box–Behnken design (BBD) techniques. 2022. South African Journal of
Chemical Engineering, 40: 87–99.
42. Veljković, V.B., Veličković, A.V., Avramović, J.M. and Stamenković, O.S. Modeling of biodiesel production:
Performance comparison of Box–Behnken, face central composite and full factorial design. 2019. Chinese Journal of
Chemical Engineering, 27(7): 1690–1698.
43. Prajapati, N., Kodgire, P. and Kachhwaha, S.S. Comparison of RSM-based FFD and CCD methods for biodiesel
production using microwave technique. 2022. Materials Today: Proceedings, 62: 6985–6991.
44. Li, T., Yang, J., Jiang, P., AlAteah, A.H., Alsubeai, A., Alfares, A.M. and Sufian, M. Predicting high-strength concrete’s
compressive strength: A comparative study of artificial neural networks, adaptive neuro-fuzzy inference system, and
response surface methodology. 2024. Materials, 17(18): 4533.
45. Raza, Y., Raza, H., Ahmed, A., Quazi, M.M., Jamshaid, M., Anwar, M.T. and Soudagar, M.E.M. Integration of response
surface methodology (RSM), machine learning (ML), and artificial intelligence (AI) for enhancing properties of polymeric
nanocomposites: A review. 2025. Polymer Composites, in press.

558