2023 International Conference on Recent Advances in Science & Engineering Technology (ICRASET)

Design and Development of Smart Farming using

ML and IoT in India
2023 International Conference on Recent Advances in Science and Engineering Technology (ICRASET) | 979-8-3503-0692-7/23/$31.00 ©2023 IEEE | DOI: 10.1109/ICRASET59632.2023.10420372

Alakananda Chatterjee Harini Viswanathan

Computer Science Engineering Department Computer Science Engineering Department
SRM Institute Of Science And Technology SRM Institute Of Science And Technology
Tamil Nadu, India Tamil Nadu, India
ac3080@[Link] hv3098@[Link]

Varsha Giridharan Akila

Computer Science Engineering Department Computer Science Engineering Department
SRM Institute Of Science And Technology SRM Institute of Science And Technology
Tamil Nadu, India Tamil Nadu, India
vg2747@[Link] akilak@[Link]

Abstract— IoT is used to find massive amounts of data This research paper utilized the Random Forest
whereas ML easily identifies trends and patterns and gives technique to analyze agricultural data and predict crop
more insights into the data that we want for farming. Focuses production. Random Forest (RF) is a versatile and efficient
on improving the productivity of farming in many aspects by machine learning method widely employed for estimating
building a recommendation model trained by supervised and
global and regional crop yields. Extensive studies have
unsupervised model ML algorithms for smart greenhouse
agricultural systems and optimizes work done by farmers as demonstrated that RF offers superior accuracy and
it involves no human intervention. REES52 IoT kit will be precision compared to other methods such as multiple
used for overall implementation. The proposed model linear regression (MLR). It is user-friendly and facilitates
employs logistic regression and a k-means algorithm to data analysis, making it the most effective approach for
predict the kind of crop that will grow followed by deploying crop yield prediction. The model developed in this study
a flask module that will help display the results user-friendly utilized authentic statistics from Tamil Nadu [2].
through a web application.
In this paper, They set up a system using machine
Keywords— REES52 IoT kit, ML algorithms, flask module literacy to prognosticate agrarian products grounded on
climatic factors. Making husbandry a successful business
I. INTRODUCTION and soliciting growers to continue crop-producing
conditioning are the only ways to end the issue. Growers
Since our everyday lives see such rapid economic
used to read their products grounded on yield data from
expansion, planning for agriculture is crucial. Numerous
previous times.
studies are being conducted on several important subjects,
including soil nutrients, crop forecasting, agricultural Therefore, there are several methods or algorithms for
practices, and crop monitoring in contemporary this kind of data analysis in crop fertilization, and these
agriculture. The key to producing high-quality crops that algorithms can be used to predict plant yields. As a result,
allow farmers to forecast crop production depending on this exploration puts forth a conception to read the crop
soil moisture is crop prediction and crop monitoring. and yield of the crop grounded on meteorological factors
Forecasting parameters like temperature, humidity, rainfall and former crop data. The planter will look at the Crop
etcetera are included in crop yield predictions and crop Advisor website, which has been designed to be stoner-
yield based on soil moisture includes a few measurements friendly and estimate how climatic factors will affect crop
like NPK (nitrogen, phosphorous, potassium) and pH products. The C4.5 algorithm is used to determine the
readings utilizing different sensors. Machine learning climatic factor that has the topmost impact on the agrarian
(ML) is a valuable decision-making technique to estimate yields of a particular crop [3].
crop yields and decide what to plant.
In this paper, They examine the most recent research
on the application of sophisticated data translation
II. LITERATURE SURVEY techniques to agriculture, particularly the development of
In a study report, writers looked at machine learning palatable grains. The framework provided in this study
techniques. The suggested strategy will advise the maps each activity specified in the stages of rice
optimum crop based on a few predicted soil features. They production and post-result, in addition to the dossier
preprocessed the data after collecting it in the form of a handled in data collecting and machine intelligence
dataset. They employed the technique known as feature approaches. They describe the data gathered and also went
engineering to raise the quality of their output. Then, they into detail about the role of machine intelligence
used machine learning methods to train and test their algorithms in farming by looking at the uses of artificial
model, with the Gradient Boosting algorithm proving to be intelligence in various scenarios, such as smart watering
the most accurate of the bunch [1]. for rice, forecasting edible grain yield estimates, tracking

979-8-3503-0692-7/23/$31.00 ©2023 IEEE

Authorized licensed use limited to: Alliance University. Downloaded on December 04,2025 at 08:04:21 UTC from IEEE Xplore. Restrictions apply.
edible grain progress, listening to edible grain diseases, In this treatise, IoT technology enables an automated
identifying the type of rice, and classifying edible grain agriculture system to track and manage key parameters like
samples [4]. temperature, humidity, soil moisture content, and sunshine.
Sensors are installed in appropriate locations, transferring
In this paper, They have secondhand electronics such
information via cloud computing to farmers' mobile
as IoT, flying fine arts, multispectral, hyperspectral, NIR,
phones. The single-board Node MCU microcontroller
thermal, RGB, and machine intelligence approaches. These
controls the system, assisting farmers in accurately
are just some of the control mechanisms considered in this
regulating irrigation systems [9].
research item that is used to mechanize farming.
Automation utilizing progressive control strategies for In this discourse, Precision Agriculture (PA) is an
agricultural practices has proven to boost crop production administration approach that exploits new and innovative
and invigorate soil fertility [5]. technologies to survive farms efficiently. The projected
order is characterized by allure talent to support farmers in
This article introduces various aspects of technologies
crop administration utilizing an IOT-based accuracy
related to agriculture. They mentioned the network
breeding platform accompanying the REES52 provisions
technologies deployed in agriculture based on IoT. This
as the IOT device and machine intelligence algorithms for
includes network architecture and layers, network
guessing [10].
topology, and protocols used. In addition, the connection
of IoT-based agricultural systems with related technologies
such as cloud computing, big data storage, and analytics III. PROPOSED SYSTEM
was also introduced. A list of smartphone and sensor-based ML helps smart farming by recommending the most
applications developed for various aspects of farm suitable crop required inside of the particular state to grow
management was also presented. In order to standardize based on top of the climatic conditions and soil nutrients
IoT-based agriculture, numerous nations have released which is given inside of the data at the same time that
legislation and recommendations. A few of these standards input. That is going to belong to the data is being analyzed
include the introduction of specific best practices. The IoT by finding out the statistics that belong to each crop based
agriculture topic was then given several open research upon the columns given the dataset, comparing the average
questions and difficulties. [6]. requirements that are going to belong to each crop with
average conditions such as during the same time that
In this research, They suggest a self-systematized
nitrogen, phosphorus, potassium, ph level, temperature,
cluster-located networking solution to defeat the
humidity, as well as the amount that belongs to rainfall
socializing for professional or personal gain-connected
inside of each state. Inside of comparison with those crops
troubles of drone-based IoT. For drone-located intelligent
which require more rainfall, temperature, humidity,
IoT, A hybrid self-systematized clustering scheme (HSCS)
nitrogen, potassium, phosphorus, as well as ph values to
is provided by them which makes use of a composite
those crops which require less. Based on top of this, the
mechanism of nocturnal luminescent beetle swarm growth
data helps us to find the best suitable crop required that is
(GSO) and bug treasure (DA). The method involves
going to belong to summer, winter, as well as rainy
choosing a cluster setup machine and a cluster head at the
seasons by using the k-means algorithm by forming
GSO site. This provides a trustworthy approach for
clusters.
following cluster appendages established by the observable
analysis of DA that guarantees good cluster administration. Using logistic regression, we are building up and
training the model that is going to belong to making the
The approaches to rescue clusters using dead cluster
model to recommend the best suitable crop required to
connection detection methods enhance network cohesion.
grow inside of each state based upon the conditions given
For HSCS, an additional overpowering procedure is
inside of the dataset. The tested model is being deployed
submitted, at which point the route draught function is
inside of an existing web application that is going to
used to choose the next leap neighbor for dossier transfer,
belong to making the model user-friendly using the Flask
guaranteeing productive communication. With an existent
framework.
composite biography-stimulated assembling algorithm,
operating HSCS is determined by conditions of cluster Using the Arduino platform as well as the REES52 IoT
building opportunity, energy devouring, cluster durability, kit which has components such during the same time that
and the tendency of transmittal success [7]. Arduino Uno, Uno cable,65pcs jumper wires,83 points
breadboard, dht11 sensor, 1 channel 5v relay module,
In this treatise, A new study on IoT-located precision
silicon techno labs LCD 16x2 alphanumeric display
farming focuses on crafty, strength-efficient dossier
(jhd162a), as well as a soil moisture sensor, helps inside of
collection at the base station. The center of authority is a
finding out moisture level inside of the soil by predicting
force-forced device that demands solar radiation to assert
on the possibility that the soil happens to be dry or wet as
energy balance. The study plans a singular production
well as we are going to find out the temperature and
density model that calculates the center of authority's
humidity by using dht11 sensor.
strength needs, which increases the adeptness of the
Improved Duty Cycling algorithm compared to additional
algorithms [8].

Authorized licensed use limited to: Alliance University. Downloaded on December 04,2025 at 08:04:21 UTC from IEEE Xplore. Restrictions apply.
 IV. ARCHITECTURE DIAGRAM B. Pseudo code for Arduino
void stup
{
// initialize serial communication at 9600 bits per
second: [Link](9600);
}
// the loop routine runs over and over again
forever: void loop()
{
// read the input on analog pin0:
int sv=analogRead(A0);
// print out the value you read:
Fig. 1. Flowchart [Link](sensor Value);
delay(1);
V. DATA COLLECTION AND PRE-PROCESSING // delay in between reads for stability
Data collection is done through pre-cleaned, pre- }
existing datasets that are available in Kaggle. A group of
discrete values describes facts, statistics, quantity, quality, VI. DEVELOPMENT OF MODEL
and other fundamental units of crops that are then further
comprehended. Before approaching data analysis, a dataset A. DESCRIPTION OF FEATURES IN DATASET
must be cleaned by fixing or deleting errors. Exploratory The proposed system is characterized by its ability to
Data Analysis performs statistical and logical techniques to support farmers in crop management using an IoT-based
analyze the basic information about the data. Data precision farming platform with the REES52 kit as the
Visualization is a graphical representation to represent the IoT device and machine intelligence algorithms for
data by comparing features in the dataset. prediction. In the machine learning approach, the dataset
helps in suggesting the crop required to grow in a
A. Pseudocode for Machine Learning particular climatic condition. The features involved are
Step 1: Start namely: N (Nitrogen), P (Phosphorous), K (Potassium),
pH, temperature, humidity, and rainfall. N quantifies the
Step 2: Input Packages soil's nitrogen content, P quantifies its phosphorous
Step 3: Load Dataset content, and K quantifies the soil's potassium level.
Step 4: Read Dataset However, temperature and humidity are inversely
Step 5: Checking statistics of crops using mean() proportional to each other and rainfall is detected based
upon temperature and humidity. Temperature is
Sum of values of all the items / total no of items measured in degrees Celsius and relative Humidity is
Step 6: Compare average requirement for each crop with measured based on the percentage value.
different conditions
Step 7: Clustering Analysis B. Benefits of Machine learning Algorithms
Amenda=[] The number of clusters depends on the distance from th
e centroid of each cluster. Data points closest to the particu
for i in range(4, 7):
larized centroid are organized into the same classification.
km=kmeans(xc=i, init=’m++’, max=500, K-Method assembling is usually second hand in retail
f_init=50, rs=0) separation, pattern acknowledgement, and figure
[Link](z) condensation.
[Link]([Link]) Predictive models that are prepared accompanying und
km=kmeans(xc=6, init=’m++’, eviating regression use records and enumerations to call co
max=500, f_init=50, rs=0) nsequences. Predictive models that are trained with linear
y_means=km.fit_predict(z) regression use records and statistics to predict outcomes.
Regression analysis is a method for learning the
Step 8: Predictive Model
connection between independent variables and dependent
model=LogisticRegression(solver=’liblinear’) variables. With discrete data, classification methods are
[Link](x_train, y_train) applied. It is used as a machine learning predictive
y_pred=[Link](x_test) modelling technique that uses algorithms to predict
Step 9: Use Flask for web application deployment outcomes in different situations. Logistic regression is
deployed to establish the relationship between various
Step 10: End
environmental variables such as temperature, precipitation,
and yield. It is important to measure the possible outcome
of the yield index and farmers will benefit from the

Authorized licensed use limited to: Alliance University. Downloaded on December 04,2025 at 08:04:21 UTC from IEEE Xplore. Restrictions apply.
outcome of this prediction. The proposed intelligent accurately suggesting the crop required to be grown in a
farming system model can provide farmers with accurate particular area by soil analysis. The temperature, humidity,
information on the crops and soil of the lands belonging to and soil moisture level values can be monitored in a serial
different states. monitor using the Arduino platform.

C. Usage of Functions And Algorithms VII. EXPERIMENTAL ANALYSIS

The data inputted in the form of a CSV file is read
using the read() function. To display the different types of
crops present in the dataset value_counts() function is
used. The functions like min() function, mean() function,
and max() are used to find the statistics of each crop
mentioned in the dataset. Comparing The average
requirement for each crop involving average conditions is
calculated using the mean() function. Various hidden
patterns are displayed in the form of graphs and arrays
about crop requirements. The dataset has been visualized
by comparing the columns in the dataset and depicting
them in the form of graphs.
The correlation matrix helps in understanding the
relationship between various variables used or the columns
Fig. 2. Elbow Graph
present in the dataset. K-means clustering mainly focuses
on the optimum number of clusters within the dataset. The This graph is obtained after using unsupervised
procedure consists of plotting the explained variability as a machine learning which involves a K-means clustering
function of the number of clusters and choosing the arc of algorithm for determining an optimum number of clusters
the curve as the number of clusters to use. The dataset has of crops based on the seasons given within the dataset by
been split into training and testing sets for validating results using “The ElbowMethod”.
and a predictive model has been created using logistic
regression. The model performance has been evaluated by
plotting a confusion matrix with actual labels (y-axis) and
predicted labels (x-axis). All 22 classes have high precision
and recall values which show that the predictive model
works very well.

D. Deployment Of Flask Framework

The compact web framework of Python is used namely
"flask”. Flask is deployed so that the trained model can be
displayed in the web application and the user can provide Fig. 3. Results after K-means Clustering Analysis
the values for each parameter shown in the web application
for predicting the best suitable crop required to grow. The This above output is obtained after applying
web application is deployed on the server for users to run. fit_predict() for performing predictions over the training
data in the form of clusters based on the columns available
E. Using IoT In Arduino Platform in the dataset.
The project aims to make an energy-efficient and
environmentally friendly grid based on precision farming.
This is achieved by measuring the temperature and
humidity of the plant growth environment using the
DHT11 sensor, which is a very inexpensive, simple digital
temperature and humidity sensor. It is characterized by low
power consumption and long-term stability. It results in
achieving relatively high measurement accuracy at a low
and affordable price. Furthermore, it also has four pins.
Applications of these sensors are measuring humidity and
temperature values in heating, ventilation, and air
conditioning systems. The humidity sensor is used as a
preventive measure in homes where people are exposed to
moisture. The number of various nutrients required for the
suggested crop to be grown is calculated using various
functions. This REES52 kit is an IOT device that helps in
Fig. 4. Confusion Matrix for Logistic Regression
the classification of soil and sensors such as DHT11 for
detecting the temperature and humidity which will help in

Authorized licensed use limited to: Alliance University. Downloaded on December 04,2025 at 08:04:21 UTC from IEEE Xplore. Restrictions apply.
 This confusion matrix represents the prediction
summary in a matrix form. The above diagram evaluates
the classification model performance using Logistic
Regression because, through this matrix, we can
understand where the classification model has gone wrong
and also guide it.

Fig. 6. Model Predicting Crop Suggestion

Fig. 5. Classification Report

The above manufacturing shows the categorization

report of a categorization model. The amount is show
accuracy, which is used to find the veracity of positive
guesses; recall, which is used to find the portion of a still
picture taken with a camera right recognized; and F1
worth, which is the mean of two together accuracy and
recall, with 1.0 recorded as the highest rank worth and 0.0
as calamity profit. Support is the number of real instances
of the class in the training basic document file of the Fig. 7. User Output
prepared categorization model. It is used to measure the
character of the indicator of the categorization invention,
namely, in our case, secondhand logistic reversion. The
manufacturing above shows by means of what many
indicators are valid and in what way or manner many are
wrong. In particular, true helpful, valid negative, wrong
certain, and dishonest negative principles are used to
forecast the rank report version.
The above outputs suggest the crop required for the
given climatic conditions to grow by testing the model
which was trained by Logistic Regression. Here the model
suggests the crop based on the values given in each feature
or column in a dataset. During a model’s testing, the values
given in decimals in the first dataset can also be rounded
off accordingly so that the model can give the most
accurate predictions. Fig. 8. Physical Demo
The values of nitrogen, phosphorus, potassium, The elements secondhand in this project are an
temperature, humidity, pH, rainfall, and the state have been Arduino Uno, a USB wire, and a jumper cable cord. Two
entered by the user. The above output predicts the kind of lengths of the cord are connected to the soil dampness
crop to be grown specified based on the distinct values. sensor by two jumper cables. The VCC fix for the soil
This application has been developed using the Flask liquid sensor is related to +5V on the Arduino Uno. The
framework. soil liquid sensor's GND is related to the Arduino Uno's
ground, and the allure parallel harvest attach is related to

Authorized licensed use limited to: Alliance University. Downloaded on December 04,2025 at 08:04:21 UTC from IEEE Xplore. Restrictions apply.
the Arduino Uno's A1 attach. A USB cord connects the the result was good. P: This identifies all instances of
Arduino Uno to your desktop computer. positive data in the dataset. By separating the total number
of helpful accidents by the number of real still pictures
taken with a device, ability determines the recall. It
subtracts the total of wrong contradictions and true still
pictures taken with a device from the number of valid still
pictures taken with a device. This method draws attention
to the fact that recall may also be thought of as the
proportion of properly detected positives to all positive
cases, which includes both genuine positives and positives
that the model missed (false positives).
F1 = 2 * PRE * REC / PRE + REC (3)
Precision: This abbreviation stands for the model's
precision, which assesses how accurately positive
predictions are made. The model's recall rate, also known
as the true positive rate, or Recall, assesses the model's
capacity to recognize positive events. The F1 score is
determined by multiplying the accuracy and recall product
by two and dividing the result by the total of the two using
the formula. Precision and recall are combined into one
statistic called the F1 score. It is especially helpful in
circumstances when balancing the trade-off between
accuracy and memory is necessary since both are crucial.

IX. CONCLUSION
Fig. 9. Output In the light of research, the ML and IoT tools will help
in the smart farming process. Achieving the goal is to
The output shows the values displayed in the serial deploy most ML and IoT solutions to help farmers with
monitor of the Arduino Uno platform which displays the soil analysis and yield forecasting. Machine Learning
soil’s moisture levels. Low moisture content is discovered predicts, yields that require specific Indian climate
in the sample soil. conditions using techniques such as K-mean clustering
with the Elbow method, logistic regression, and learn and
VIII. EQUATIONS test values. The IoT gadget called REES52 includes
PRE = TP / TP+FP (1) sensors like the soil moisture sensor and a simple and
extremely affordable digital temperature and humidity
How many of the occurrences that were favorably sensor. Measures ambient air using a thermistor and a
expected and turned out to be real positives is a measure of capacitive humidity sensor.
precision, or how accurate the positive predictions were?
We know the number of cases that the model properly REFERENCES
identified as positive as TP. These are the instances where
[1] Shafiulla Shariff, Shwetha R B, Ramya O G, Pushpa H, Pooja K
the model called for a favorable outcome and it came about R, Crop Recommendation using Machine Learning
favorably. False Positives: This statistic shows they Techniques,(August 18, 2022). Available:
misclassify how many occasions the model is positive. [Link]
These are the situations where the model expected a learning-techniques
favorable result, but the result was unfavorable. [2] P. Priya, [Link] and [Link], Predicting Yield Of
The Crop Using Machine Learning Algorithm,(April 2018).
By separating the total number of valid positives per
[3] [Link], Bharat Misra, CD Singh, Machine learning
picture taken with a camera into true and dishonest still approach for forecasting crop yield based on climatic
pictures, the precision is deliberate. parameters,(October 16, 2014), Available:
[Link]
REC = TPR = TP / P = TP / FN + TP (2) [4] R. Alfred, J. H. Obit, C. P. -Y. Chin, H. Haviluddin and Y. Lim,
Towards Paddy Rice Smart Farming: A Review on Big Data,
Recall and true positive rate, two related variables, are Machine Learning, and Rice Production Tasks, IEEE Access, vol.
represented in the formula above. These two measures 9, pp. 50358-50380, 2021.
assess how well a classification algorithm performs in [5] Syeda Iqra Hassan, Muhammad Mansoor Alam, Usman Illahi,
accurately recognizing positive events. The number of Mohammed A. AL Ghamdi, Sultan H. Almotiri, and Mazliham
Mohd Suud, A Systematic Review on Monitoring and Control
cases that the model properly identified as positive is Strategies in Smart Agriculture, (February 8, 2021), Available:
known as TP. These are the instances where the model [Link]
called for a favorable outcome and it came about [6] Muhammad Shoaib Farooq, Shamyla Riaz, Adnan Abid, Kamran
favorably. False Negative is a measure of how many Abid, and Muhammad Azhar Naeem, “A Survey on the Role of
occurrences the model misclassified as negative. These are IoT in Agriculture for the Implementation of Smart
Farming”,(October 25, 2019).
the situations where the model projected a bad result, but
[7] Farooq Aftab, Ali Khan, Zhongshan Zhang,” Hybrid Self-

Authorized licensed use limited to: Alliance University. Downloaded on December 04,2025 at 08:04:21 UTC from IEEE Xplore. Restrictions apply.
 Organized Clustering Scheme for Drone Based Cognitive Internet [9] Bammidi Deepa, Chukka Anusha, and P. Chaya Devi, “Smart
of Things”,(April 29, 2019). Available: Agriculture Using IoT”, (August 11, 2020).
[Link] [10] Ritesh Kumar Singh, Rafael Berkvens, Maarten Weyn,
[8] Himanshu Agrawal, Ruchi Dhall, K.S.S Iyer and Vijayalakshmi “Agrifusion: An Architecture for IoT and Emerging Technologies
Chetlapalli, “An improved energy efficient system for enabled Based on a Precision Agriculture Survey”, (September 29, 2021).
precision agriculture”, (June 18, 2019). Available: [Link]

Authorized licensed use limited to: Alliance University. Downloaded on December 04,2025 at 08:04:21 UTC from IEEE Xplore. Restrictions apply.