Smart Financial Decision-Making System

Assel Mukasheva Almas Saduakas Alibek Bisembayev

School of Information Technology and School of Information Technology and School of Information Technology and
Engineering Engineering Engineering
2025 IEEE/IEIE International Conference on Consumer Electronics-Asia (ICCE-Asia) | 979-8-3315-7402-4/25/$31.00 ©2025 IEEE | DOI: 10.1109/ICCE-Asia67487.2025.11263733

Kazakh-British Technical University Kazakh-British Technical University Kazakh-British Technical University

Almaty, Kazakhstan Almaty, Kazakhstan Almaty, Kazakhstan
mukashevascience@[Link] [Link]@[Link] [Link]@[Link]

Dina Koishiyeva Jeong Won Kang

School of Information Technology and Department of Transportation System
Engineering Engineering
Kazakh-British Technical University Korea National University of
Almaty, Kazakhstan Transportation
[Link]@[Link] Uiwang-Si, Republic of Korea
jwkang@[Link]

Abstract— The integration of artificial intelligence (AI) and potential crises or fraudulent activities, and reacts
big data analytics into financial systems has paved the way for automatically through pre-defined financial workflows. By
intelligent decision-making tools. This paper introduces a Smart incorporating RSI into its analytics engine, the system
Financial Decision-Making System designed to support enhances its ability to assess the momentum of financial
individuals and institutions in managing investment, credit, and
savings decisions. Utilizing machine learning algorithms and
assets and evaluate conditions like overbought or oversold
predictive modeling, the system offers real-time suggestions markets. When irregularities are identified, the BPMN-
based on historical data, behavioral patterns, and economic integrated response engine can autonomously execute tasks
indicators. The model is trained on diversified financial datasets such as reallocating capital, halting risky transactions, or
to ensure accuracy, scalability, and contextual adaptation. Our notifying relevant stakeholders. Furthermore, the system’s
findings indicate that the smart system improves financial architecture is designed to be modular, scalable, and suitable
outcomes on average across multiple user groups when for deployment in real-world financial environments.
compared to traditional advisory methods. We conclude by The proposed approach advances the field by integrating
discussing future integration opportunities and challenges in predictive analytics with operational automation,
deploying such AI-driven solutions. In addition to financial
recommendation capabilities, the system is also designed to
demonstrating how a unified system can simultaneously
detect early signs of crises, fraudulent behavior, and abrupt interpret financial signals and initiate intelligent responses.
market shifts. By integrating Business Process Model and The model's effectiveness is validated through experiments
Notation (BPMN), the system can automatically respond to showing improved financial outcomes and faster reactions to
anomalies—such as reallocating funds or notifying relevant changing market conditions.
departments. It leverages time series analysis and over 50 years
of historical asset data, including stock and commodity prices, II. RELATED WORKS
to enhance the reliability of forecasts and support strategic Financial decision-making remains a fundamental
financial planning.
component of achieving success at both personal and
Keywords—Machine learning algorithms, time series, industrial levels. Yet, the fast-paced evolution of global
forecasting stock market prices. markets, the increasing sophistication of financial
instruments, and the vast growth of data have rendered
I. INTRODUCTION traditional decision-making approaches increasingly
Financial decision-making remains a fundamental ineffective. To address these challenges, there is growing
component of achieving success at both personal and momentum behind intelligent financial systems that leverage
industrial levels. Yet, the fast-paced evolution of global artificial intelligence (AI), technical analysis, and process
markets, the increasing sophistication of financial automation to deliver real-time, adaptive insights.
instruments, and the vast growth of data have rendered This paper introduces a Smart Financial Decision-Making
traditional decision-making approaches increasingly System that combines machine learning techniques, decades
ineffective. To address these challenges, there is growing of historical asset data, technical indicators such as the
momentum behind intelligent financial systems that leverage Relative Strength Index (RSI), and automated decision logic
artificial intelligence (AI), technical analysis, and process based on Business Process Model and Notation (BPMN). The
automation to deliver real-time, adaptive insights. system is not limited to offering static financial advice; it also
This paper introduces a Smart Financial Decision-Making proactively identifies unusual market behaviors, detects
System that combines machine learning techniques, decades potential crises or fraudulent activities, and reacts
of historical asset data, technical indicators such as the automatically through pre-defined financial workflows. By
Relative Strength Index (RSI), and automated decision logic incorporating RSI into its analytics engine, the system
based on Business Process Model and Notation (BPMN). The enhances its ability to assess the momentum of financial
system is not limited to offering static financial advice; it also assets and evaluate conditions like overbought or oversold
proactively identifies unusual market behaviors, detects markets. When irregularities are identified, the BPMN-

Authorized licensed use limited to: The Oxford College of Engineering. Downloaded on December 11,2025 at 06:14:28 UTC from IEEE Xplore. Restrictions apply.
integrated response engine can autonomously execute tasks At the heart of this architecture is a database that serves as a
such as reallocating capital, halting risky transactions, or common foundation. The new structure covers two main
notifying relevant stakeholders. Furthermore, the system’s areas. The first is the integration of a knowledge base and a
architecture is designed to be modular, scalable, and suitable model base, which together provide information support for
for deployment in real-world financial environments. The decision-making by combining qualitative analysis (based on
proposed approach advances the field by integrating logical inference) and quantitative analysis (based on model
predictive analytics with operational automation, calculations). The second area involves the use of data
demonstrating how a unified system can simultaneously warehouses and intelligent analysis technologies (data
interpret financial signals and initiate intelligent responses. mining) to extract valuable information and patterns from
The model's effectiveness is validated through experiments large data sets. The knowledge gained reflects the internal
showing improved financial outcomes and faster reactions to patterns and trends inherent in large volumes of financial
changing market conditions. data.
III. METHODOLOGY
B. Intelligent data analysis technology and its application
in financial management.
This study uses two types of data sets: source and training.
The source set, formed on the basis of a large-scale exchange-
The intelligent financial management support system uses
traded fund (ETF), is used for initial model training. The
data storage and intelligent analysis (data mining)
training dataset is used at the stage of solving new problems.
technologies, applying relational storage formats to process
During training, the source data undergoes pre-processing
large volumes of business information and increase the
and is fed into a pre-trained model. The architecture of the
scalability of system management. Parallel computing
financial decision support system has been designed with
methods are used to process complex queries, which
scalability, modularity and adaptability in mind. It integrates
optimises queries to support decision-making. Based on
historical data analysis, machine learning algorithms and
extensive corporate data, the system generates flexible and
technical indicators, including the relative strength index
interactive statistical trend analyses and forecast reports using
(RSI) and MACD, and provides for automated process
mathematical models and data processing methods.
execution using BPMN. This multi-level architecture ensures
not only high accuracy of financial forecasts, but also
intelligent response to dynamic changes in market In addition, the system implements various intelligent
conditions[1]. analysis models that automatically process data, perform
inductive reasoning, and establish relationships between
information in the data warehouse. These models allow you
A. Architecture of an intelligent financial management
to identify hidden patterns and potential templates that are
system.
key to forecasting and supporting management decisions,
which helps to form new business models. Ultimately, this
An intelligent decision support system (IDSS) is formed by helps managers develop effective market strategies and make
integrating a decision support system (DSS) and an expert informed decisions. The system architecture includes three
system (ES). DSS mainly includes a model database, a levels: the data collection level, the data storage and
human-machine interaction system, and a database, while ES organisation level, and the analysis and visualisation level, as
consists of an inference engine, a dynamic database, and a shown in Figure 2.
knowledge base. The general structure of an intelligent
financial management support system combining these two
components is shown in Figure 1.

Fig . 2. Research framework for intelligent management using data mining

Fig . 1. System architecture illustration of the intelligent financial
management support platform

Authorized licensed use limited to: The Oxford College of Engineering. Downloaded on December 11,2025 at 06:14:28 UTC from IEEE Xplore. Restrictions apply.
IV. STUDY OF INTELLIGENT DATA ANALYSIS ALGORITHMS IN lowest or measure level, which encompasses alternative
FINANCIAL MANAGEMENT SYSTEMS schemes and strategies that can be adopted to fulfill the
objective. Although the hierarchy itself provides a useful
Determining the optimal sample splitting method requires a representation of the relationships between elements across
function that allows you to establish which split provides the different levels, it is not sufficient to solve complex decision-
best classification balance. The information gain indicator is making problems. To address this limitation, AHP relies on
used for this purpose. In an intelligent data analysis the construction of pairwise comparison matrices, where
algorithm, when selecting attributes as a test feature for the elements are evaluated against one another to determine their
current node, the one that provides the maximum increase in relative importance. Through this process, the method
information is selected. This approach minimises uncertainty quantifies the degree of preference between attributes and
in classification and builds the simplest and most informative generates weight coefficients that support the prioritization of
decision tree that reflects the data structure. decision alternatives.

Further mathematical justification is presented below. The AHP also integrates predictive modeling as part of its broader
ID3 (Iterative Dichotomiser 3) algorithm is based on a analytical framework. While classification and estimation
procedure for selecting features for decision tree nodes, in focus on existing conditions, prediction extends the analysis
which information gain is the main criterion. Each non-leaf to future outcomes. By constructing trend models based on
node is checked, and the tree is split according to the attribute current data, AHP enables the forecasting of future values of
with the greatest increase in information. This process relevant attributes. Since many datasets exhibit cyclical or
continues until all subsets contain objects of only one class. traceable patterns, the development of appropriate predictive
The result is a decision tree that can effectively classify new models allows for the anticipation of long-term trends.
examples. Practical applications include forecasting enterprise financial
returns, predicting quarterly product demand, and estimating
The method for calculating information gain is as follows: the the future price dynamics of production resources[3].
information gain for each attribute is calculated and
compared. The attribute with the greatest gain is selected as Another important component of intelligent financial
the division criterion. Let there be a sample S. If the number analysis supported by AHP is association analysis, which
of different values of class attribute p is m, then the situation identifies relationships and rules among data objects. This
where all m values belong to one type is defined as follows: technique is widely employed in market basket analysis,
where correlations in sales data are examined to inform
𝑇(𝑠1 ⋯ 𝑠𝑚 ) = ∑𝑚 () product placement, promotion strategies, and distribution
𝑖=1 𝑃𝑖 log 2 𝑃𝑖
planning. However, its utility extends far beyond commerce,
Which represents the entropy of a sample set S, where m as it can also reveal causal patterns in broader organizational
denotes the number of distinct categories (class labels) and 𝑃𝑖 and financial processes. Building on these methods,
corresponds to the probability of occurrence of class i. This enterprise financial analysis systems based on data mining are
measure, derived from information theory, quantifies the designed to serve executives, financial managers, and
degree of uncertainty or impurity within the dataset. In the analysts engaged in intelligent financial management. By
context of the ID3 algorithm, entropy serves as the basis for investigating and clarifying user requirements, these systems
calculating information gain, enabling the selection of identify conceptual relationships among potential entities
attributes that most effectively partition the data and enhance within functional modules. This process provides a strong
the accuracy of decision tree classification.[2] foundation for the subsequent development of logical and
physical database structures. Such systems typically
The Analytic Hierarchy Process (AHP) has become widely encompass a range of functions, including enterprise
applied in intelligent financial management systems due to its financial management, asset inventory and allocation,
ability to combine both objective and subjective factors depreciation and verification, and asset maintenance. The
within a clear hierarchical structure. Unlike approaches based entity–relationship (E-R) model is particularly effective in
solely on subjective judgment, AHP begins with an initial capturing and representing the relationships between entities
subjective determination of relative weights between and attributes, thereby ensuring that the conceptual structure
indicators and then incorporates mathematical derivation and of the system is both coherent and operationally useful[15].
quantification. This integration ensures that the results
obtained are more scientific and accurate, while still V. RESULTS
reflecting the complexity of decision-making processes.
The dataset utilized in this research was sourced from the
The methodology of AHP involves decomposing complex Bloomberg database provided by Refinitiv, which delivers an
problems into a hierarchical structure where elements are extensive and precise compilation of financial information
grouped according to their attributes and arranged across encompassing the equity, bond, fund, and foreign exchange
different levels based on dominance relationships. Typically, markets. From this source, more than thirty publicly listed
three levels are distinguished within this structure: the highest firms were randomly chosen, and their trading data were
level, which defines the overall objective or desired outcome; retrieved for the period spanning 2018 to 2021. During the
the intermediate or criterion level, which represents the set of preprocessing stage, it was discovered that several stocks
rules and factors necessary to achieve the target; and the included incomplete transaction records, with certain

Authorized licensed use limited to: The Oxford College of Engineering. Downloaded on December 11,2025 at 06:14:28 UTC from IEEE Xplore. Restrictions apply.
companies having been introduced to the market only in The analysis shows that the first improved version of the
2011. These incomplete cases were excluded, and the dataset algorithm consumes more time than the baseline under
was subsequently refined. Because only the daily opening certain support conditions. Nevertheless, when evaluating the
prices were accessible, additional preprocessing steps were average across the entire sequence of runs, the optimized
necessary to calculate the daily rate of return, defined as the algorithm consistently surpasses the baseline in efficiency.
difference between the closing price of the current day and The most refined implementation demonstrates markedly
that of the previous day, divided by the previous day’s closing superior performance, affirming that the applied
price. Following this adjustment, stocks with positive daily improvements significantly strengthen computational
returns were classified as exhibiting growth, while negative effectiveness.
values indicated declining performance. In its final form, the
dataset comprised 25 stocks and a total of 6,100 entries, the To provide further confirmation of the optimized fuzzy
structure of which is presented in Figure 3. model’s validity, comparative experiments were conducted
with a BP neural network and a multiple linear regression
(MLR) framework. MLR represents a standard multivariate
statistical approach that establishes relationships between
independent and dependent variables through the fitting of
observed data using a linear equation, thereby extending
simple regression methods. The BP neural network, by
contrast, is a prominent technique within machine learning
that is extensively applied across disciplines. A typical BP
architecture is organized into three layers: an input layer that
receives external data, a hidden layer responsible for internal
computations, and an output layer that produces results until
convergence to the desired value is achieved.

VI. Discussion and Future Work

The presented Smart Financial Decision-Making System

Fig . 3. Data structure of the intelligent financial management system
demonstrates the effectiveness of integrating artificial
intelligence, predictive modeling, and BPMN-based
automation for identifying market anomalies and optimizing
To assess the performance of the enhanced algorithm, the financial strategies. As part of future research, we plan to
processed data were first subjected to the baseline method extend these methodological principles to other domains
with the minimum support threshold set to 0.01. Typically, a characterized by complex market dynamics and decision
smaller support value generates a larger collection of uncertainty. One promising direction involves the application
candidate sets, thereby increasing computational complexity of the proposed algorithms to the development of an
but offering a stronger foundation for comparative analysis. electronic livestock exchange - a digital simulation platform
As the minimum support value rises, the execution time of designed to model and evaluate the breeding and economic
the baseline, the initially modified, and the further optimized potential of animals. In this system, machine learning
algorithms tends to decline. Lower thresholds result in a algorithms and behavioral data analytics can be employed to
higher frequency of candidate itemsets and necessitate estimate the virtual value of livestock based on phenotype
additional comparisons, which in turn elevates the runtime. A characteristics, genetic evaluation indexes, and user-defined
comparative depiction of the execution times across the three economic priorities. By comparing the AI-generated virtual
algorithms under varying support thresholds is illustrated in prices with actual market data, it becomes possible to
Figure 4. calibrate the weighting structure of the economic index and
identify discrepancies between expected and real valuations.
This approach mirrors the adaptive feedback loop
implemented in financial decision models and can contribute
to more accurate trend forecasting prior to real-world
auctions. Integrating the algorithms developed in this
research, particularly those for anomaly detection, time series
forecasting, and automated BPMN-based responses will
allow the virtual exchange to function as a decision-support
ecosystem for the agricultural sector. In doing so, it will
provide not only market foresight but also enhance
sustainability through intelligent resource allocation and
early detection of inefficiencies in breeding processes.

VII CONCLUSION
This paper introduced an intelligent financial decision-
Fig . 4. Temporal comparison chart making framework that combines time series prediction and

Authorized licensed use limited to: The Oxford College of Engineering. Downloaded on December 11,2025 at 06:14:28 UTC from IEEE Xplore. Restrictions apply.
technical analysis. This work advances a solution predicated [3] C. Xiang and B. Li, “Research on ship intelligent manufacturing data
monitoring and quality control system based on industrial internet of
on recent progress in object-oriented software engineering, things,” The International Journal of Advanced Manufacturing
through which a new class of decision-support systems can Technology, vol. 107, no. 3-4, pp. 983–992, 2020.
be realized. Contemporary developments in software [4] C. Geng, Y. Xu, and N. Metawa, “Intelligent financial decision support
architecture, design patterns, and methodological practice system based on data mining,” Journal of Intelligent and Fuzzy
have introduced fresh conceptual frames and technical Systems, vol. 2, pp. 1–10, 2021.
techniques; taken together, these technologies furnish more [5] L. Jia, “Intelligent data mining algorithm application in enterprise
financial management strategy optimization: a Ecommerce
convenient tooling for building intelligent decision-support perspective,” International Journal of Security and its Applications,
systems (IDSS), compress the software development vol. 10, no. 7, pp. 157–168, 2016
lifecycle, and strengthen system scalability and extensibility. [6] Dinesh, S., Rahul, S., Sandeep, O., Relangi, K. and Rama Raju, A.M.S.,
The study first reviews the research status of classical 2021. Stock price prediction using LSTM. Mukt Shabd Journal, 10(6),
pp.436-442.
decision-support systems (DSS) and intelligent decision-
[7] Lin, Z., 2018. Modelling and forecasting the stock market volatility of
support systems, outlining associated challenges, and then SSE Composite Index using GARCH models. Future Generation
examines their structural models and defining characteristics. Computer Systems, 79, pp.960-972.
Drawing on pertinent theoretical results, it specifies the [8] Y. Lu,“Financial accounting intelligence management of internet of
foundational concepts and properties of an intelligent things enterprises based on data mining algorithm,” Journal of
financial management support system, diagnoses design Intelligent and Fuzzy Systems, vol. 37, no. 5, pp. 5915–5923, 2019.
deficiencies, and proposes an enterprise-grade system [9] C. Jun and P. Jin, “Financial management and decision of decision tree
algorithm based on data mining,” IPPTA: Quarterly Journal of Indian
architecture for intelligent financial management grounded in Pulp and Paper Technical Association, vol. 30, no. 8, pp. 70–74, 2018.
data-mining technologies. The exposition then introduces [10] Q. Wang and B. Zhang, “Research and implementation of the
essential data-mining preliminaries—definitions, end-to-end customer-oriented modern hotel management system using fuzzy
process, and algorithmic families—after which an analytical analytic hiererchical process (FAHP),” Journal of Intelligent and Fuzzy
synthesis of an existing “intelligent” data-mining algorithm Systems, vol. 40, no. 4, pp. 8277–8285, 2021.
exposes its limitations. An improved algorithm is [11] Y. Cui, “Intelligent recommendation system based on mathematical
modeling in personalized data mining,” Mathematical Problems in
consequently formulated to address these shortcomings and Engineering, vol. 2021, no. 3, Article ID 6672036, 11 pages, 2021.
is assessed via targeted mining experiments; the empirical [12] Niyazova, K., Mukasheva, A., Balbayev, G., Iliev, T., Mirambayeva,
findings indicate that the refined method offers measurable N., & Uzakbayev, M. (2024). Ant colony optimization algorithm for
advantages over the baseline. feature selection in suspicious transaction detection
system. Engineering Proceedings, 60(1), 18.
ACKNOWLEDGMENT [13] Mukasheva, A., Saparkhojayev, N., Akanov, Z., Apon, A., & Kalra, S.
(2019). Forecasting the prevalence of diabetes mellitus using
This research was funded by the Science Committee of the econometric models. Diabetes Therapy, 10(6), 2079-2093.
Ministry of Science and Higher Education of the Republic of [14] S. Lessmann, B. Baesens, H. V. Seow, and L. C. Thomas (2015)
Benchmarking state-of-the-art classification algorithms for credit
Kazakhstan (grant no. BR28712579). scoring: an update of research, *European Journal of Operational
Research*, 247(1), pp. 124–136.
REFERENCES [15] H. Sun, Z. Yao, and Q. Miao (2021) Design of macroeconomic growth
prediction algorithm based on data mining, *Mobile Information
[1] X. Gong, Z. Wang, and L. Wang, “Research on financial early warning Systems*, 2021(7), Article ID 2472373, 138 pages.
model for papermaking Enterprise based on particle swarm K-means
algorithm,” Paper Asia, vol. 34, no. 6, pp. 41–45, 2018.
[2] L. Wang, Y. Liu, and J. Wu, “Research on financial advertisement
personalised recommendation method based on customer
segmentation,” International Journal of Wireless and Mobile
Computing, vol. 14, no. 1, pp. 97–121, 2018.

Authorized licensed use limited to: The Oxford College of Engineering. Downloaded on December 11,2025 at 06:14:28 UTC from IEEE Xplore. Restrictions apply.