Source-linked AI summary
Data Mining Applications: A comparative Study for Predicting Student's performance
Surjeet Kumar Yadav, Brijesh Bharadwaj, Saurabh Pal
TL;DR
The paper addresses the use of data mining for extracting knowledge from educational data and predicting student performance. It applies decision-tree classification to past student-performance information, evaluates common tree algorithms, and reports CART as the best classifier in the study. The approach is intended to support earlier identification of students needing special attention.
Problem
Educational institutions have substantial student data but limited use of knowledge-discovery approaches for educational decision making and performance prediction.
Method
The study applies decision-tree classification to attendance, class-test, seminar, and assignment data to predict end-semester student performance.
Results
CART is reported as the best algorithm for classification of the student data among the frequently used decision-tree classifiers studied.
Takeaways & Limitations
The study is intended to help identify students needing special attention and support appropriate action before the next semester examination.
Abstract
from arXiv · showhide
Knowledge Discovery and Data Mining (KDD) is a multidisciplinary area focusing upon methodologies for extracting useful knowledge from data and there are several useful KDD tools to extracting the knowledge. This knowledge can be used to increase the quality of education. But educational institution does not use any knowledge discovery process approach on these data. Data mining can be used for decision making in educational system. A decision tree classifier is one of the most widely used supervised learning methods used for data exploration based on divide & conquer technique. This paper discusses use of decision trees in educational data mining. Decision tree algorithms are applied on students' past performance data to generate the model and this model can be used to predict the students' performance. It helps earlier in identifying the dropouts and students who need special attention and allow the teacher to provide appropriate advising/counseling.
I. INTRODUCTION
The paper frames educational data mining as a way to extract useful knowledge from student data and study academic performance. It applies decision-tree classification to student information and reviews related studies linking student characteristics and educational data to performance.
- Motivation: Student academic achievement matters to educators, administrators, employers, and broader social and economic development.The paper identifies CGPA as a measure of overall academic performance across university examinations.
- Educational Data Mining: Knowledge discovery and data mining are presented as methods for extracting useful patterns from large educational data repositories to support decision making.Educational data mining applies techniques including decision trees, neural networks, Naïve Bayes, and K-nearest neighbour methods.
- Study Objective: The paper uses classification and decision trees to predict end-semester performance from attendance, class-test, seminar, and assignment marks.It investigates the accuracy of different decision-tree algorithms using student information collected from a management system.
- Related Work: Related educational data-mining work also used enrollment or student data to classify success, predict results, and warn students at risk before final examinations.One cited case study reported accuracies of 59.4 for CHAID and 60.5 for CART.
- Related Work: Prior studies used decision trees, Bayesian classification, regression, clustering, and association rules to examine student performance, learning behavior, and educational choices.Reported predictors include academic background, family income, mother’s education, living location, teaching medium, and other student characteristics.
III. DECISION TREE INTRODUCTION
Decision trees are introduced as interpretable classifiers built through growth and pruning phases. The paper focuses its experiments on ID3, C4.5, and CART, using 10-fold cross-validation to assess classification accuracy.
- Decision Tree Structure: A decision tree represents attribute tests at internal nodes, outcome-labeled arcs, and class labels at leaf nodes.The structure is described as a flow-chart-like tree with internal nodes and leaves.
- Growth and Pruning: Tree construction has a growth phase that recursively splits training data until partitions predominantly contain records from the same class.The growth phase can overfit the training data.
- Growth and Pruning: Pruning removes noise, outliers, and unnecessary complexity from a fully grown tree to address overfitting and increase classification accuracy.Pruning requires less time than building the tree, while growth requires multiple passes over the training data.
- Algorithms and Evaluation: The experiments use ID3, C4.5, and CART because these are identified as the most frequently used decision-tree algorithms.The paper evaluates them on student data using classification procedures.
A. ID3 (Iterative Dichotomiser 3)
ID3 selects splits using information gain and can handle continuous attributes through discretization or thresholding. Unlike C4.5, it does not support pruning.
- Algorithm Overview: ID3 is a decision-tree algorithm introduced by Quinlan Ross and based on Hunt’s algorithm.Its tree construction consists of building and pruning phases in the general description.
- Attribute Selection: ID3 chooses the splitting attribute with the highest information gain and uses that attribute as the root node.Possible attribute values become arcs, and instances are recursively tested for class membership.
- Attribute Handling: ID3 handles continuous attributes by discretizing them or finding a threshold that provides a suitable split.The algorithm can use attribute values directly when determining a best split point.
- Pruning and Comparison: ID3 does not support pruning, whereas C4.5 uses pessimistic pruning to remove unnecessary branches and improve classification accuracy.C4.5 also uses gain ratio and supports categorical and continuous attributes.
C. CART
CART builds decision trees for categorical and continuous attributes, handles missing values, and produces binary splits. It uses Gini Index selection and cost-complexity pruning.
- Algorithm Overview: CART stands for Classification And Regression Trees and was introduced by Breiman.The algorithm is based on Hunt’s algorithm.
- Attribute Handling: CART handles categorical and continuous attributes and missing values when building a decision tree.These capabilities are described as part of CART’s tree-construction procedure.
- Splitting and Selection: CART uses Gini Index for attribute selection and produces binary trees through binary splits.Unlike ID3 and C4.5, its Gini Index does not use probabilistic assumptions.
- Pruning: CART applies cost-complexity pruning to remove unreliable branches and improve classification accuracy.The pruning mechanism is presented as a way to reduce unreliable parts of the tree.
IV. DATA MINING PROCESS
Student performance is determined through internal assessment and the end-semester examination, with internal assessment based on several educational activities.
- Internal assessment uses class tests, seminars, assignments, general proficiency, attendance, and lab work to evaluate students.
- The end-semester examination provides a separate score, and students must achieve minimum marks to pass the semester.
A. Data Preparations
The study prepares student records by joining and cleaning database tables, selecting mining variables, and defining response and predictor categories.
- Data Preparations: Records from different tables were joined into one table, and errors were removed before analysis.
- Data Preparations: The study selected required fields and derived variables, with predictor and response variables documented in Table II.
- Data Preparations: Class-test grades were categorized as Poor, Average, or Good using thresholds below 40%, from 40% to below 60%, and at least 60%.
- Data Preparations: Seminar performance was classified by presentation and communication skills, while assignments were coded according to submission.
- Data Preparations: Attendance was categorized as Poor below 60%, Average from 60% to below 80%, or Good at least 80%; 70% is normally required for examination participation.
- Data Preparations: Lab work was represented as Yes when completed and No when not completed.
- Data Preparations: End semester marks were defined as the response variable and divided into First, Second, Third, and Fail classes using percentage thresholds.
C. Data Set
The dataset contains records for 48 MCA students from VBS Purvanchal University’s Computer Applications department across sessions 2008 to 2011.
- C. Data Set: The study used a dataset of 48 students from the MCA course at VBS Purvanchal University, Jaunpur, Uttar Pradesh.
- C. Data Set: The records cover sessions from 2008 to 2011 and are associated with the Computer Applications department.
D. Model Construction
Model construction used Weka to apply and evaluate decision-tree algorithms on the prepared dataset, including ID3, C4.5, and CART.
- D. Model Construction: Weka provides graphical tools for data preprocessing, classification, regression, clustering, association, visualization, and feature selection.
- D. Model Construction: Weka’s Classify panel applies algorithms, estimates predictive-model accuracy, and visualizes erroneous predictions or the model.
- D. Model Construction: The study used the ID3, C4.5, and CART algorithms for classification from the decision-tree algorithms implemented in Weka.
E. Results Obtained
The study reports classifier accuracy results for ID3, C4.5, and CART, alongside model-building execution time and classification matrices.
- Table IV reports the accuracy of ID3, C4.5, and CART using 10-fold cross-validation.
- The results section includes execution time measurements for building each classifier model.
- Separate classification matrices report prediction-model results for ID3, C4.5, and CART.
- The paper also presents figures comparing classifiers and showing rules generated by a decision tree.
V. CONCLUSION
The conclusion presents decision trees as interpretable classifiers and identifies CART as the best-performing algorithm in the experiments. It connects the study to improving student performance and identifying students needing special attention.
- Decision trees are valued because their classification rules are easier to interpret than those of other methods.
- The experiments found CART to be the best algorithm for classifying the student data.
- The study is intended to help teachers and students improve student performance.
- It is also intended to identify students needing special attention and support action for the next semester examination.