Evaluation of School Students Performance Using Machine Learning

N. T. Renukadevi, K. Saraswathi, E. Roshini, M. G. Lakshitha and S. Pratheeksha

Department of CT-UG, Kongu Engineering College, Erode, Tamil Nadu, India

Keywords:

Support Vector Machines, Decision Trees, Random Forest, K-Nearest Neighbour, Naïve Bayes.

Abstract: In today's educational perspective, the need for data-driven insights to enhance student outcomes is

increasingly recognized. In this paper we are going to develop a machine learning model to predict and

evaluate student performance based on various academic and demographic factors. This system will utilize

past information about students like their grades and background to provide educators better suggestions on

how to assist students in choosing their academic group for the 11th grade. The dataset will be prepared by

collecting information from school students through Google forms. To predict and evaluate student’s

performance, we apply four distinct machine learning models like Decision Trees, Random Forest, Support

Vector Machines (SVM), and Logistic Regression. This research exposes the application of machine

learning in guiding the selection of academic groups with promising results and significant potential in

educational settings. Furthermore, the research underscores the importance of using data-driven approaches

to support educators in making informed decisions and also this can assist in altering personalized

interventions enhancing learning outcomes for all students.

1 INTRODUCTION

In today's fast-changing world of education, there's a

growing recognition that using data to improve

student outcomes is essential. Traditional methods

of guiding students may not be enough to address

the varied and complex needs of modern learners.

Schools are trying to support student success more

effectively and advanced technologies like ML are

emerging as promising tools. This paper discusses a

detailed study on how MLmodels developed and

predict student performance, particularly to help

students choose their academic groups, specifically

for 11th grade. This is done by analyzing the

academic and personal factors, like student's grades

and socio-economic backgrounds and goals to give

educators strong, data-based recommendations to

better support students in making informed decisions

about their future studies.

To succeed in this research, a dataset will be

created by collecting information from students

through Google Forms. This ensures that the data

may vary and reflects the many different aspects that

can show a student's performance. The data will be

analyzed using four different ML methods: Random

Forest, K-Nearest Neighbours, Naïve Bayes, and

Adaptive Boosting. Each of these methods has its

own strengths in recognizing patterns and making

accuracy, which will allow for a comparison of how

to evaluate student performance and select academic

groups.

The research aims to show how machine learning

can predict the student's performance by using

historical data. This not only makes more accurate

predictions about student performance but also helps

educators for more personalized educational support

for each student, leading them to choose academic

groups for better learning outcomes. This study

highlights the value of using data-driven approaches

in education. This gives educators an actionable

analysis based on real data; machine learning can

significantly improve the decision-making processes

involved in academics. Additionally, these

technologies can help make educational practices,

ensuring that all students get the guidance and

support to succeed academically.

Thus, this research explores how machine

learning can be used in education, predicting a new

way to evaluate student performance and help the

educators to guide the students in academic group

selection. The promising results from these models

suggest that machine learning has the potential to

provide educational practices, providing a new level

of support for both educators and students. Thus, as

Renukadevi, N. T., Saraswathi, K., Roshini, E., Lakshitha, M. G. and Pratheeksha, S.

Evaluation of School Students Performance Using Machine Learning.

DOI: 10.5220/0013622900004664

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 3rd International Conference on Futuristic Technology (INCOFT 2025) - Volume 3, pages 503-512

ISBN: 978-989-758-763-4

503

we continue to explore all these technologies, it

becomes clear that data-driven approaches are

necessary steps in improving educational outcomes

for all students.

ML is a part of AI that focuses on developing

methods for computers to learn from data and make

predictions or decisions. The process starts with

collecting data, which can come from various

sources like databases, text, or images. This data is

then pre-processed to ensure whether the data is

ready for analysis. Once the data is prepared,

machine learning algorithms are applied. These

algorithms can handle different types of problems:

supervised learning is used when the model learns

from labeled data to predict outcomes, while

unsupervised learning is useful for finding patterns

in data that is not labeled. There's also reinforcement

learning, which trains models to make decisions in a

step-by-step manner mainly used in robotics and

gaming. The model is trained on historical data to

recognize patterns and can be used to predict new

data. To ensure the working of the model, it is

evaluated using specific metrics and often adjusted

or fine-tuned to improve the performance. The final

step involves comparing different algorithms to find

the best for the problem. Machine learning's ability

has made it useful in many areas like healthcare,

finance, understanding language, and recognizing

images etc., It helps organizations make smarter

decisions based on data.

2 LITERATURE REVIEW

Esmael Ahmed (Ahmed, 2024) conducted a

comparative study on the challenges faced by

contemporary educational institutions in analyzing

the efficiency; endowing sound education,

formulating strategies for evaluating students’

performance and recognizes future needs. In this

study the data is collected from the Wollo University

learning management system and grouped together

to predict the student’s end outcome based on their

demographic details and performance in entrance

exams, and with various factors etc., In this paper,

performance of the classifiers such as SVM, DT,

NB, and KNN are examined. It is observed that

SVM algorithm provides better outcomes with 96%

accuracy.

In research work, done by P. Krishna Reddy et

al. (Reddy, Bavankumar, et al. , 2024) the data is

collected with 22 attributes. This study applied an

empirical method to opt for students’ records and

decide on important variables for analysis. The study

has applied five different data mining techniques

PAC, SVM, LDA, RNC and ET and subsequently

compares the consequences of five ML algorithms to

identify the paramount performing algorithm. And

from this research SVM provides the highest

accuracy of 94.86%.

Various studies have addressed the challenge of

predicting student performance using machine

learning techniques; Nitin Yadav et al. used UCI

machinery student dataset as input and pre-

processed the data to select the attributes. The study

solved this problem by applying SVM, NB, C4.5,

ID3 and found that SVM gives more accuracy of

88%. For future work, select the best attribute sets

like Stud_Name, Gender, Previous Exam Marks,

Address, Parents Education, etc. with ML algorithm

to provide the high accuracy result of prediction of

student performance system (Yadav and Deshmukh,

2023).

Another comparative study was done by Yawen

Chenand and Linbo Zhai(Chen and Zhai, 2023).

First the three datasets were selected which are

oriented to Student performance, Engineering

placements prediction and Student admission and

also done pre-processing to select the specific

features. In this subsection, the study applies seven

popular ML algorithms including KNN, DT, RF,

LR, SVM, NB, and ANN. RF provides highest

accuracy among those seven ML algorithms.

Nuha Alruwais and Mohammed Zakariah

(Alruwais and Zakariah, 2023) said that assessing

acquaintance of student is crucial to determine

student progress and providing feedback to improve

student performance.Thisprovides feedback to

improve student performance. The study used 2

different data sets, the dataset 1 includes all the

attributes and the dataset 2 removes the least

correlated variables to create a smaller dataset. To

solve this problem, the 7 different classifiers are

used, including SVM, LR, RF, DT, GBM, GNB, and

MLP. Then the performance of Dataset 1 and

Dataset 2 were compared. This resulted that the

GBM exhibited the highest prediction accuracy of

98%.

Rosemary Vargheese et al. (Vargheese, Peraira,

et al., 2022) proposes Student Performance Analysis

System (SPAS) to remain track of students’ results.

During the implementation phase, to generate rules

for prediction of students’ performance, they

analyzed 114 students’ records by using data mining

techniques for their future. This research has 6

applied four different data mining techniques SVM,

DT, RF, NB. After comparison, it is concluded that

the SVM has provided the highest accuracy

INCOFT 2025 - International Conference on Futuristic Technology

504

(81.82%) among other classification techniques.

They predicted the performance of the students in

two phases (i) training phase, and (ii) testing phase.

The current prediction model is not dynamically

updated within the system's source code. For future,

they introduce a dynamic prediction model allowing

that model to be retrained automatically whenever

new training data is added to the system with ML

algorithm for the prediction of the student

performance system.

Jovana Jovic et al. Educational Data Mining has

conducted an EDM approach in order to classify and

predict student performance with ML techniques.

ML algorithms such as LR, LDA, KNN, DT, NB

and SVM have been used. The research classified

that SVM shows the best results with accuracy of

88.5%. Thus, the future work will analyze a larger

number of ML algorithms and try to gain a more

accurate model for the prediction of student

performance and support learning (Jović, Kisić, et

al. , 2022).

Educational data mining has emerged as a

powerful tool for predicting students' academic

success and uncovering hidden relationships in

educational data. Mustafa Yagci (Yağcı, 2022)

presented a novel model in this study that uses

machine learning (ML) algorithms to forecast

undergraduate students' final exam grades. The

midterm exam grades of the students are the source

data for this project. The final exam grades of the

students were predicted by calculating and

comparing the performances of the RF, SVM, LR,

NB, and KNN. In the future, it may be possible to

review students' working methods and enhance their

performance by projecting their achievement grades.

AlabbasHayderassesses and contrasts the

accuracy, precision, recall, and prediction of two

machine learning algorithms—SVM and ANN—

when trained to classify binary datasets. Three

distinct subquestions were used to break down each

research question. In this study, the seven

researchers developed two machine learning models

and evaluated their efficacy. An ANN model was the

second, and an RF model was the first. Future

research could look at utilizing each course's degree

separately rather than just the course pass rate to

determine how each course's performance impacts

future academic results. They used ANN, SVM and

other ML classification algorithms for the prediction

of students’ performance (Hayder, 2022).

Charalampos Dervenis et al. (Dervenis,

Kyriatzis, et al. , 2015) has completed a project on

learning analytics which measures, gathers and

analysis data about learners and their contexts in

order to optimize learning and the environments in

which it takes place. This represents a tangible step

toward a more change in a society that is increased

by algorithms. The algorithms such as KNN, RF and

SVM were tested. The test data is used to gauge the

algorithm's performance, training data served as the

foundation for its creation. Every data set is run

through the test data and outcome is predicted. The

RF produced the beat results then followed by KNN

and SVM. The study made accurate prediction for

passes or fails using the model. For future behaviors,

identify potential problems at an early stage or even

improve inter-institutional collaboration and develop

an agenda for the larger community of students and

teachers.

Opeyemi Ojajuni et al. (Ojajuni, et al. , 2021) has

done a study using ML to explore and to predict

student academic performance by analyzing data

from 1,044 students, including demographic, social,

and academic information. It applies various

supervised ML algorithms, such as RF, SVM, GBM,

DT, LR, and DL, to classify and predict outcomes.

The research involves data preprocessing and feature

engineering, categorizing final grades into excellent,

good, satisfactory, poor, and failure. XGBoost

showed the highest accuracy of 97.12%. The study

suggests that future research could investigate

additional ML models and address challenges like

overfitting and model deployment in real

educational settings.

Poonam Sawant et al. (Sawant, Gupta, et al. ,

2021) examine the impact of the COVID-19

pandemic on student performance in India,

highlighting the challenges faced due to the shift

from offline to online learning. Various ML

algorithms, such as DT and NB, are used to analyze

and predict student performance during this period.

Findings indicate that students' performance

decreased during the pandemic, with NB showing

better accuracy than the DT. Future research aims to

incorporate more advanced algorithms to enhance

the performance evaluation system.

Samah Fakhri Aziz et al. explored using ML to

predict student performance. It compares Gradient

Boosted Decision Trees, RF, DT, and DL

algorithms. This paper showed that DL is the highest

accuracy in predicting student performance. The

study used a dataset of 450 students with 17

attributes, and the pre-processing involved checking

and handling missing values. Overall, DL was found

to be the best for predicting how well students will

do (Aziz, 2020).

Ali Salah Hashim et al. (Hashim, Awadh, et al. ,

2020) evaluated number of supervised ML

Evaluation of School Students Performance Using Machine Learning

505

algorithms such as DT, NB, LR, SVM, KNN,

Optimization and NeuralNetwork had performed in

predicting student’s performance on final exams.

The results showed that the logistic regression gave

most accurate accuracy as 68.7% for passed students

and 88.8% for failed students. Data pre-processing

involves data preparing, combining and cleaning the

data in order to train them was the first step. The

subject of second step is the most frequently used

algorithm’s classification performance according to

ML technique.

J. Dhilipan et al. (Dhilipan, Vijayalakshmi, et al.

, 2021) explores the application of ML techniques to

forecast academic success in educational settings.

The study utilizes various algorithms, including

Binomial Logistic Regression, DT, Entropy, and

KNN, to analyze student performance based on their

10th, 12th, and semester marks. In this study The

Binomial Logistic Regression has achieved the

highest accuracy with a rate of 97.05%. The results

demonstrate that these methods can help educators

identify students at risk of underperforming and

guide them in improving academic outcomes

Rajalaxmi et al. (Rajalaxmi, Natesan, et al. , 2019).

The research suggests that in future additional

features are added to our dataset to acquire better

accuracy Saraswathi et al (Saraswathi, Renukadevi,

et al. , 2024).

3 METHODOLOGIES USED

3.1 K-NearestNeighbors

KNN is a supervised learning algorithm used for

both classification and regression tasks. Data Points

are classified or predicted according on the

maximum number ofclasses.The "k" value denotes

the number of neighbors for the classification or

regression decision.Distance metrics commonly

used include Euclidean distance, whichmeasures the

similarity between data points, based on the

Euclidean distance calculation shown

in Equation (1).KNN is a lazy-learner because it

does not make assumptions based on underlying

data and defers computation until predictions are

needed. KNN is computationally expensive for large

datasets, and its performance can be affected by

noise, and its role extends to areas such as

recommendation systems and anomaly detection.

)()(

2121

yyxx

−−

(1)

KNN is known for its simplicity and

intuitiveness, with no training phase required. It can

handle non-linear decision boundaries and complex

patterns without making strong assumptions about

data distribution to make it suitable for multiclass

problems. However, KNN is a sensitive choice to

the hyper parameter "k". Though it calculates

distance for all data points it is computationally

expensive for large data sets. It can suffer from the

"curse of dimensionality", because when the number

of features increases the performance will decrease.

The diagrammatic representation is shown in Figure

Figure 1: K - Nearest Neighbors

3.1.1 How KNN Works

Step 1: Determine the value of “k”.

Step 2: Calculate the distance of all data points from

existing data using the Euclidean distance measuring

technique.

Step 3: Choosing the number of nearest data points to the

new data based on the “k” value.

Step 4: For each data point, one of its K neighbors predicts

the class.

Step 5: Counting the data points according to the class

labels and resulting the majority voted label to the new

data point.

3.2 Decision Tree

DT is a supervised machine learning algorithm used

for both classification and regression tasks. It splits

the dataset into subsets based on the most important

feature, by creating a tree-like structure as shown in

Figure 2, where each internal node represents a

decision based on feature, each branch fits to an

outcome of that decision, and each leaf node

represents the final classification. The algorithm

selects the best features for splitting by maximizing

information gain by calculating entropy for each as

defined by equation (2). DT are easy to visualize and

INCOFT 2025 - International Conference on Futuristic Technology

506

useful for understanding the decision-making

process. RF oftenleverages multiple DT for

improved accuracy.

)(log))((

pipiSHE



(2)

DT is transparent and easy to understand, offering

human-readable decision rules. They handle both

classification and regression tasks. DT is robust to

outliers and alsoaccommodates missing value. They

are exposed to overfitting, especially when the tree

is deep and complex, it will be sensitive to small

changes in the data which leads to different tree

structures. DT may struggle with imbalanced

datasets and are limited in capturing decision

boundaries compared to ensemble methods.

Additionally, the greedy search during tree

construction may not always lead to globally

optimal solutions Renukadevi et al..

Figure 2: Decision Tree

3.2.1 Decision Tree analysis

Step 1: Find the problem.

Step 2: Build thestructure of the decision tree.

Step 3: Identify the alternatives of the decision.

Step 4: Calculate entropy for each feature.

Step 5: Use entropy to calculate information gain.

Step 6: Discover the possible outcomes.

Step 7:Examine the best decision.

3.3 Support Vector Machines

SVM is a supervised machine learning algorithm

that can be applied to classification and regression

problems. It functions by locating the hyperplane in

a high dimensional space that divides data points

into distinct classes the best, as defined by equation

(3). The data points that are closest to the hyperplane

are known as “support vectors”, and they have an

impact on the hyperplane's orientation and position

which is shown in Figure 3. The goal of SVM is to

maximize the margin, or the separation between

each class's closest data points and the hyperplane.

The margin is calculated as defined in the equation

(4).With the use of kernel functions, the algorithm

can be expanded to handle non-linear relationships

and is efficient when processing high-dimensional

data. SVM is renowned for its reliable operation and

capacity to manage intricate decision boundaries

across a range of applications.

0. =+ bxw

(3)

||||

arg

inM =

(4)

SVM is effective at finding optimal hyper planes

for binary and multi-class classification, and also

uses kernel functions to handle both linear and non-

linear decision boundaries. They are robust to

outliers and offer high accuracy when properly

tuned, making them suitable for complex and high-

dimensional datasets.SVMs however, can be

computationally demanding for large datasets and

have limited interpretability, particularly when

dealing with non-linear kernels. They are also prone

to overfitting if not properly regularized and can

struggle with noisy or overlapping classes.

Figure 3: Support Vector Classifier

3.3.1 Steps in the SVM Algorithm

Input Data:Gatherlabeled training data using class

labels and feature vectors.

Linear SVM:Choose the optimal hyperplane if

the data can be divided linearly.

Evaluation of School Students Performance Using Machine Learning

507

Non-Linear SVM:If the data cannot be separated

linearly,use the kernel trick to convert it into a

higher dimension.

Optimal Hyperplane:Determine which

hyperplane maximizes the difference in class

margins..

Optimization Problem:The process of finding the

best model parameters, hyperparameters, or features

to minimize or maximize the performance metrics,

such as error, accuracy or loss.

Support Vectors:To define the hyperplane, find

the closest support vectors.

Prediction:Determine which side of the

hyperplane new data falls on and classify it

accordingly.

3.4 Random Forest

RF is an ensemble learning technique based on DT.

It builds multiple decision trees during training and

outputs the classification or regression as defined in

the equation (5) of the individual trees for a more

robust and accurate result. The key idea is to

introduce randomness in the tree-building process by

considering a random subset of features and data

points for each tree. This helps mitigate over-fitting

and enhances the model's generalization ability. RF

is known for its high predictive accuracy, resilience

to outliers, and suitability for complex datasets. It is

widely used in various applications, Including

classification, regression, and feature importance

analysis in machine learning.





−

xTb

)(

(5)

Ensemble methods, like RF, improve predictive

accuracy and reduce overfitting by aggregating

multiple decision trees. They are robust to outliers

and noisy data which handle both classification and

regression tasks and can effectively manage high-

dimensional data and large datasets and shown in

Figure 4. Additionally, they provide feature

importance scores, which aid in feature selection and

interpretation. However, ensemble methods are less

interpretable than a single DT. They may not

perform as well as gradient boosting for certain tasks

and require more memory and computational

resources.

3.4.1 The steps to build a random forest are

•

Understand the problem: Define the

objectives

•

Understand the data: Get to know the raw

materials

•

Data preprocessing: Prepare the data

•

Data analysis: Build models

•

Results interpretation and

evaluation: Interpret and evaluate the

results

•

Data report and communication: Create a

data report and communicates the results

Figure 4. Random Forest

3.5 Naive Bayes

NB is a simple and fast ML algorithm used for

classification tasks. It is based on Bayes' theorem,

which calculates the probability of a class given in

the features of the dataset. In NB, the posterior

probability is used to calculate the probability of a

data point belonging to a particular class as defined

in the equation (6). The algorithm assumes that all

features are independent of each other, so that it is

called"naive". And also,Naive Bayes works well in

many real-world applications, especially with large

datasets.

( | , ,..., ) ( ) ( | )

CX X X PCx PX C

∏

(6)

NB is fast and efficient, especially with large

datasets and it performs well for tasks like text

classification and spam detection. It’s easy to

implement and works well with a small amount of

data. However, the main limitation is the assumption

that all features are independent, which may not hold

true values in many cases, potentially reducing

accuracy. It also struggles when features are closely

related.

INCOFT 2025 - International Conference on Futuristic Technology

508

Figure 4: Naive Bayes Diagram

3.5.1 The steps of the Naive Bayes classifier

are

Training: Use the training data to estimate the

parameters of a probability distribution.

Prediction: For new test data, calculate the

posterior probability of that sample belonging to

each class.

Classification: Classify the test data according to

the largest posteriorprobability.

3.6 Boosting Algorithm

Boosting is a machine learning technique that

improves the accuracy of models by combining the

strengths of multiple weak models to create a

stronger overall model. It works by training models

sequentially, where each new model focuses on

correcting the errors made by the previous ones.

3.6.1 Common types of boosting algorithms

include

Ada-Boost -This adjusts the weights of incorrectly

classified data points.

Gradient Boosting - This optimizes performance

by minimizing errors through gradient descent.

XGBoost–isan efficient version of Gradient

Boosting known for its speed and performance on

large datasets.

Boosting algorithms are powerful for improving

model accuracy by combining multiple weak models

into a strong one. They are effective for complex

datasets and can handle various types of data.

However, boosting can be sensitive to noisy data

and outliers which can lead to overfitting if not

carefully managed. It also requires more

computational resources and time due to its

sequential nature and tuning the model can be

complex.

Figure 5. Boosting Algorithm

3.7 Gradient Boosting

Gradient Boosting is a popular boosting algorithm

that combines multiple weak models to create a

strong predictive model. It works by iteratively

training decision trees to predict the residual errors

of the previous iteration, using gradient descent to

minimize the loss function. The process starts with

an initial simple model, and then, at each iteration, a

new decision tree is trained to predict the difference

between the observed output and the predicted

output from the previous iteration. The predictions

from each tree are weighted and combined to update

the overall model. The gradient descent step adjusts

the weights to minimize the loss function, typically

mean squared error or logistic loss, ensuring that

each subsequent tree focuses on correcting the errors

of the previous ones. This sequential process

continues until a specified number of trees is

reached or a stoppingcriterion is met, resulting in a

highly accurate and robust predictive model.

In this system, we are predicting and evaluating

the performance of students based on their academic

factors using ML algorithms. Block diagram is

shown in Figure 6. We have collected data from the

school students, upto 170 data for our project.Our

research mainly focuses to support educators in

making informed decisions and also assist in altering

personalized interventions enhancing learning

outcomes for all students.

Figure 6. Block Diagram

Evaluation of School Students Performance Using Machine Learning

509

4 RESULTS AND DISCUSSION

In this research work, various ML algorithms were

implemented for predicting the performance of the

student. We collect the information from the school

students through google form. The dataset was split

into two parts: 80% for training the model and 20%

for testing it. This is a common practice in machine

learning to make sure the model works well with

new data.

4.1 Environment Setup

The special tools and programs were used to conduct

the experimentation, including Google Colab for

data visualization, and Microsoft Excel for data

handling. Python was chosen due to its easy-to-learn

syntax and the availability of libraries like Numpy,

Pandas, Scikit-learn, and Sklearn. Numpy calculates

mean values; Pandas fetches data from files, creates

data frames, and handles data frames. Scikit-learn,

also known as Sklearn, contains machine learning

tools like classification and so on.

4.2 Data Pre-Processing

The dataset was cleaned by removing missing

values, extracting features and selecting key features

25. Here is an elaboration on each step:

•

Removing Missing Values - Data often has

missing or NAN values, It will affect the

accuracy of the dataset

•

There are several techniques for handling

missing data such as:

•

Removal - If the amount of missing values

is small, rows or columns with missing

values can be dropped.

•

Imputation - Missing values can be filled in

with statistical measures such as the mean,

median, or mode, or even using more

sophisticated methods like regression or

interpolation.

•

Feature extraction – It involves the process

of creating new features from existing data.

These new features can have better patterns

in data and it will improve the model

performance.

There are different techniques it includes:

Time-related Features - If you have date-time

data, you can extract features like the day of the

week, month, or hour.

Textual Features - For text data, you might

extract features like word counts, or use techniques

like TF-IDF or word embeddings.

Statistical Features - Generating features based

on statistical properties like mean, variance, or

correlation can add valuable information.

4.2.1 Selecting Key Features

Correlation Analysis - It identifies the features and it

is highly related with the target variable. It helps to

find the outcome more.

Variance Threshold - Low variance features do

not help in providing any useful information.

Feature Importance - Algorithms like decision

trees or random forests can ranked as a importance

feature, because it will help to improve the

prediction by showing which algorithm is useful for

the model.

Dimensionality Reduction – In this process we

eliminate the unwanted features without losing much

information. We can easily improve the accuracy By

using these steps, the dataset is prepared for analysis

or model building, resulting in a cleaned data.

In this study, we applied several Machine

Learning algorithms, including KNN, Decision Tree

(DT), SVM, Naive Bayes (NB), Gradient Boosting

(GB), and Random Forest Classifier (RFC), to

predict student performance based on data collected

from school student. Here we use factors such as

Q_TOTAL, H_TOTAL, AVERAGE as the feature

selection subset and RESULT is the target to predict

the accuracy of the algorithm. The dataset includes

170 students, with their average scores ranging from

198.5 to 483.5.The average score across all students

is approximately 357.2with most students scoring

between 302.5 and 431.Students were categorized

into four groups based on their results are Above

Average: 59 students (average score:

394.0),Average: 49 students (average score:

317.6),Excellent: 33 students (average

score:468.8),Below Average: 28 students (average

score: 217.5). In order to choose a classifier for

predicting the final exam outcome (whether the

course was passed or failed), a 10-fold cross-

validation approach was conducted. Cross validation

approach splits the training dataset into 10 groups of

approximately equal size, trains the model on nine

groups and tests the model on the tenth group in ten

iterations. The outcomes of the experiments are

summarized using classifier accuracy that was

calculated as the average accuracy after ten cross

validation iterations.

INCOFT 2025 - International Conference on Futuristic Technology

510

Gradient boosting is a popular boosting

algorithm in machine learning used for classification

and regression tasks. Boosting is one kind of

ensemble Learning method which trains the model

sequentially and each new model tries to correct the

previous model. It combines several weak learners

into strong learners. We have used this algorithm for

comparative analysis and we that It has the accuracy

of 96%.

Table 1: Accuracy measures of ML algorithms

ALGORITHM ACCURACY

KNN 88%

DT 91%

SVM 99.23%

NB 84%

GB 96%

RFC 94%

Accuracy is a metric that measures how often a

model correctly predicts the outcome. It is the ratio

of correctly predicted instances (both true positives

and true negatives) to the total number of instances.

Accuracy is commonly used for classification tasks.

KNN gives the accuracy as 88%, In DT has the

accuracy of 91%, In SVM we have implemented

Normalization and we get the accuracy as 99.23%,

NB has the accuracy of 84%, GB has the accuracy of

88.24%, RF has the accuracy of 94%. Among the

algorithms tested, the Support Vector Machine has

the most accuracy, with an 99.23% prediction

accuracy, compared to the other models.

REFERENCES

E. Ahmed, “Student Performance Prediction Using

Machine Learning Algorithms,” Applied

Computational Intelligence and Soft Computing, vol.

2024, 2024, doi: 10.1155/2024/4067721.

P. K. Reddy, S. Bavankumar, Y. P. Mohaideen, and G.

Kishore, “Students’ performance evaluation using

machine learning algorithms”, Material Science and

Technology, 2024, vol.23, pp.310 – 322.

N. R. Yadav and S. S. Deshmukh, “Prediction of Student

Performance Using Machine Learning Techniques: A

Review,” 2023, pp. 735–741. doi: 10.2991/978-94-

6463-136-4_63.

Y. Chen and L. Zhai, “A comparative study on student

performance prediction using machine learning,” Educ

Inf Technol (Dordr), vol. 28, no. 9, pp. 12039–12057,

Sep. 2023, doi: 10.1007/s10639-023-11672-1.

N. Alruwais and M. Zakariah, “Evaluating Student

Knowledge Assessment Using Machine Learning

Techniques,” Sustainability (Switzerland), vol. 15, no.

7, Apr. 2023, doi: 10.3390/su15076229.

R. Vargheese, A. Peraira, A. Ashok, and B. Johnson,

“Students’ Performance Analysis Using Machine

Learning Algorithms”,International Journal of

Research in Engineering and Science, 2022, vol.10,

no.6, pp.1804-1809.

J. Jović, E. Kisić, M. R. Milić, D. Domazet, and K.

Chandra, “Prediction of student academic performance

using machine learning algorithms, Proccedings of the

Thirteenth International Conference on e-Learning,

2022.

M. Yağcı, “Educational data mining: prediction of

students’ academic performance using machine

learning algorithms,” Smart Learning Environments,

vol. 9, no. 1, Dec. 2022, doi: 10.1186/s40561-022-

00192-z.

Alabbas Hayder,“Predicting Student Performance Using

Machine Learning: A Comparative Study Between

Classification Algorithms” ,

DigitalaVetenskapligaArkivet, 2022.

C. Dervenis, V. Kyriatzis, S. Stoufis, and P. Fitsilis,

“Predicting Students’ Performance Using Machine

Learning Algorithms,” in ACM International

Conference Proceeding Series, Association for

Computing Machinery, Sep. 2022. doi:

10.1145/3564982.3564990.

O. Ojajuni et al., “Predicting Student Academic

Performance Using Machine Learning,” in Lecture

Notes in Computer Science (including subseries

Lecture Notes in Artificial Intelligence and Lecture

Notes in Bioinformatics), Springer Science and

Business Media Deutschland GmbH, 2021, pp. 481–

491. doi: 10.1007/978-3-030-87013-3_36.

P. Sawant, S. Gupta, Y. Sharma, and A. Singh,

“Classification Approach for Evaluating Students

Performance in Covid 19 Pandemic,” Int J Eng Adv

Technol, vol. 10, no. 4, pp. 110– 113, Apr. 2021, doi:

10.35940/ijeat.d2374.0410421.

S. F. A. Aziz, “Students’ Performance Evaluation Using

Machine Learning Algorithms,” College Of Basic

Education Research Journal, vol. 16, no. 3, pp. 977–

986, Jul. 2020, doi: 10.33899/berj.2020.166006.

A. S. Hashim, W. A. Awadh, and A. K. Hamoud, “Student

Performance Prediction Model based on Supervised

Machine Learning Algorithms,” in IOP Conference

Series: Materials Science and Engineering, IOP

Publishing Ltd, Nov. 2020. doi: 10.1088/1757-

899X/928/3/032019.

J. Dhilipan, N. Vijayalakshmi, S. Suriya, and A.

Christopher, “Prediction of Students Performance

using Machine learning,” IOP Conf Ser Mater Sci

Eng, vol. 1055, no. 1, p. 012122, Feb. 2021, doi:

10.1088/1757-899x/1055/1/012122.

Rajalaxmi, R. R., Natesan, P., Krishnamoorthy, N., &

Ponni, S. (2019). Regression model for predicting

engineering students academic

performance. International Journal of Recent

Technology and Engineering, 7(6S3), 71-75.

N. T. Renukadevi, S. Nanthitha, K. Saraswathi, S. Shobika

and R. T. Karthika, "WhatsApp Group Chat Analysis

by using Machine Learning," 2023 International

Evaluation of School Students Performance Using Machine Learning

511

Conference on Sustainable Computing and Data

Communication Systems (ICSCDS), Erode, India,

2023, pp. 340-346, doi:

10.1109/ICSCDS56580.2023.10104961.

K. Saraswathi, N. T. Renukadevi, K. G. Akshaya and S.

Kanishka, "Hyper Parameter Optimization in Machine

Learning For Enhancing Loan Sanction

Processes," 2024 15th International Conference on

Computing Communication and Networking

Technologies (ICCCNT), Kamand, India, 2024, pp. 1-

7, doi: 10.1109/ICCCNT61001.2024.10724586.

INCOFT 2025 - International Conference on Futuristic Technology

512