Source-linked AI summary
Predicting Students' Performance Using ID3 And C4.5 Classification Algorithms
Kalpesh Adhatrao, Aditya Gaykar, Amiraj Dhawan, Rohit Jha, Vipul Honrao
TL;DR
The paper addresses the need to estimate newly admitted students’ future academic performance from prior academic and admission data. It applies classification with ID3 and C4.5 to student records and reports approximately 75.275 average accuracy across bulk and singular evaluations. The system’s current scope excludes extracurricular activities and does not dynamically update prediction parameters.
Problem
Educational institutions need prior knowledge of enrolled students’ likely performance to identify promising students and support those expected to receive lower grades.
Method
The system applies ID3 and C4.5 classification to student admission records and prior academic results to predict future Pass or Fail outcomes.
Results
75.275 average percentage of accuracy was achieved for 182 students across Bulk and Singular Evaluations.
Takeaways & Limitations
The learned predictive patterns were implemented in a web application for staff members to obtain predicted results for admitted students.
Takeaways & Limitations
The current implementation does not dynamically update prediction parameters and excludes extracurricular activities and vocational courses that may affect overall performance.
Abstract
from arXiv · showhide
An educational institution needs to have an approximate prior knowledge of enrolled students to predict their performance in future academics. This helps them to identify promising students and also provides them an opportunity to pay attention to and improve those who would probably get lower grades. As a solution, we have developed a system which can predict the performance of students from their previous performances using concepts of data mining techniques under Classification. We have analyzed the data set containing information about students, such as gender, marks scored in the board examinations of classes X and XII, marks and rank in entrance examinations and results in first year of the previous batch of students. By applying the ID3 (Iterative Dichotomiser 3) and C4.5 classification algorithms on this data, we have predicted the general and individual performance of freshly admitted students in future examinations.
1. INTRODUCTION
The paper motivates using past student performance and admission information to predict future academic outcomes. It applies ID3 and C4.5 to first-year engineering student data after pruning the dataset.
- Institutional and educational-background differences make prior performance useful for estimating students’ probable future academic performance.
- The dataset includes gender, class X and XII board scores, entrance-examination scores, category, admission type, and student-identification information.
- After pruning the data, ID3 and C4.5 were applied to predict first-semester results as precisely as possible.
2. LITERATURE SURVEY
The literature survey frames classification as supervised prediction into predefined classes and distinguishes it from clustering. It also discusses missing-data handling and accuracy evaluation for this task.
- Data mining discovers patterns, associations, changes, structures, and anomalies from large data repositories for applications including machine learning and decision support.
- Classification learns rules from labelled data and assigns test data to predefined classes; here, each student is labelled “Pass” or “Fail.”
- Unlike classification, clustering discovers previously unknown classes, making it unsuitable for predicting the predefined Pass and Fail outcomes.
- Missing values can arise from deletion, non-entry, perceived unimportance, or failure to register data changes, and several replacement or manual-review strategies are available.
- The study ignores tuples with missing attributes because admission-record fields are mandatory and missing values are considered unlikely.
- Classification accuracy is the percentage of tuples assigned to the correct class, although incorrect assignments may also carry costs.
2.4. ID3 Algorithm
The section describes ID3 and C4.5 decision-tree learning, including entropy-based splitting, information gain, normalized information gain, pruning, and handling varied training data.
- ID3 Algorithm: ID3 generates a decision tree that classifies database tuples after the tree models the classification process.
- ID3 Algorithm: ID3 selects the condition attribute with the highest information gain to reduce the entropy of the resulting subsets.
- ID3 Algorithm: Entropy measures the order in a database state, with H = 0 identifying a perfectly classified set.
- ID3 Algorithm: Information gain compares the original dataset entropy with the weighted entropies of the subdivided datasets.
- C4.5: C4.5 extends ID3 by supporting missing values, differing attribute costs, post-creation pruning, and discrete or continuous attributes.
- C4.5: C4.5 recursively selects the attribute with the highest normalized information gain to split samples into increasingly smaller subsets.
3. TECHNOLOGIES USED
The implementation uses web technologies and data-management tools to support the student-performance prediction application. RapidMiner generates the ID3 and C4.5 decision trees.
- HTML and CSS: HTML structures web documents and supports embedded images, objects, and interactive forms.
- HTML and CSS: CSS separates document content from presentation, enabling shared formatting across multiple pages and reducing redundancy.
- PHP and the CodeIgniter Framework: PHP provides the server-side scripting language used for web development and can be embedded in HTML.
- PHP and the CodeIgniter Framework: CodeIgniter supplies PHP libraries and functionalities through a simple interface and was used to build the application front end.
- MySQL stores user information and student data in the web application.
- RapidMiner supports data loading, preprocessing, modelling, evaluation, and deployment, and was used to generate ID3 and C4.5 decision trees.
4. IMPLEMENTATION
The implementation was organized into five stages, from collecting and preparing student data through decision-tree processing, testing, and web-application development.
- The implementation comprised five stages covering data collection, preprocessing, decision-tree construction, testing, and front-end development.The stages are depicted in the processing model.
- Student information was collected from students admitted to the second year, including details submitted during enrolment.
- Extraneous information was removed, and relevant data was stored in a database before algorithmic processing.
- ID3 and C4.5 were applied to training data to obtain decision trees, which were then used with first-year test data.
- The final implementation stage developed the front end as a web application.
4.1. Student Database
The training dataset combined enrolment, academic, demographic, and entrance-examination information and was transferred from Excel into a MySQL database for data mining.
- The training dataset contained information about students admitted to the first year.
- Recorded attributes included gender, caste, board-examination percentages, Class XII subject percentages, entrance-examination marks, and admission type.
- The source data was provided as a Microsoft Excel 2003 spreadsheet containing individual student details.
- The student data was entered into a MySQL database to facilitate data-mining operations.
4.2. Data Preprocessing
Data preprocessing selected attributes considered relevant to performance prediction, removed extraneous fields, and converted numerical marks into discrete categories.
- The training data was segmented using feasible splitting attributes considered likely to have higher impact on student performance.Location was given as an example of a splitting attribute.
- Irrelevant fields such as residential address, name, application ID, and admission date were removed from the database.
- Retained attributes included entrance-examination merit, gender, Class XII Physics, Chemistry and Mathematics percentages, and admission type.
- A class attribute was added to hold the predicted result.
- Marks were discretized into defined classes, including “good” and “bad” merit categories based on a threshold of 120 out of 200.
4.3. Data Processing Using RapidMiner
The pruned student database was processed in RapidMiner using classification algorithms, producing decision trees from ID3 and C4.5 for evaluating student performance.
- The pruned student database was fed into RapidMiner to evaluate results from classification algorithms on the training dataset.
- ID3 produced a decision tree containing the splitting attributes.
- C4.5 also produced a decision tree from the student data.
- The C4.5 tree had fewer decision nodes than the improved ID3 tree.
4.4. Implementing the Performance Prediction Web Application
The system converted predictive patterns from ID3 and C4.5 decision trees into a web application for staff to evaluate individual or uploaded student records. It also provided verification and evaluation-history functions, with both algorithms reaching 75.145% accuracy on the reported test.
- Predictive patterns discovered with RapidMiner were implemented in a working web application for staff members.
- The application supports simultaneous staff registration and login so multiple staff members can work without interrupting one another.
- Decision trees from improved ID3 and C4.5 were translated into PHP if-else class methods using PCM percentage, merit marks, admission type, and gender.
- Staff can perform singular evaluations by entering student attributes and receive a predicted “Pass” or “Fail” result.
- Bulk Evaluation allows staff to upload spreadsheets and display each student’s predicted class, while the Verify tab compares predictions with actual results.
- 75.145% accuracy was reported for both ID3 and C4.5 on the verification test, with mismatched tuples shown for incorrectly predicted records.
5. FUTURE WORK
Future work targets dynamic updating of prediction parameters and broader student data coverage. The authors specifically identify extracurricular activities and vocational courses as omitted factors that may affect prediction accuracy.
- The current implementation does not dynamically update RapidMiner-generated decision trees when new training sets are added.
- Future development will make the implementation train its prediction parameters when new training sets are fed into the web application.
- Extracurricular activities and completed vocational courses were not considered, although the authors believe including them may improve prediction accuracy.
6. CONCLUSIONS
The system predicts first-year engineering students’ results from prior first-year results of second-year students using ID3 and C4.5. Across bulk and singular evaluations, the reported average accuracy for 182 students was approximately 75.275.
- The system predicts current first-year engineering students’ results using first-year results from students currently in the second year.
- Bulk Evaluation results are reported in Table 1, while random individual test cases produced approximately equal accuracy in Table 2.
- Approximately 75.275 average accuracy was achieved across Bulk and Singular Evaluations for 182 students.