Source-linked AI summary
A CHAID Based Performance Prediction Model in Educational Data Mining
M. Ramaswami, R. Bhaskaran
TL;DR
The paper addresses the need to predict higher secondary students’ performance from diverse influencing factors so that slow learners can be identified. It uses selected variables to construct and validate a CHAID prediction model from Indian school data. The model achieved 44.69% overall prediction accuracy, correctly classifying 345 of 772 students, and exceeded an earlier model’s accuracy.
Problem
Higher secondary performance depends on many factors, creating a need for accurate predictive models that can identify students with low academic achievement.
Method
The study selects relevant features using Pearson chi-square testing and constructs a tree-based CHAID model whose nodes split student populations by predictor values.
Results
44.69% overall model prediction accuracy was achieved, correctly classifying 345 of 772 students, higher than an earlier model by Al-Radaideh et al.
Takeaways & Limitations
Medium of instruction, secondary-school marks, school location, living area, and type of secondary education were the strongest student-performance indicators.
Takeaways & Limitations
The outcomes cannot be generalized because the sample was limited and geographically restricted to schools in districts of Tamilnadu.
Abstract
from arXiv · showhide
The performance in higher secondary school education in India is a turning point in the academic lives of all students. As this academic performance is influenced by many factors, it is essential to develop predictive data mining model for students' performance so as to identify the slow learners and study the influence of the dominant factors on their academic performance. In the present investigation, a survey cum experimental methodology was adopted to generate a database and it was constructed from a primary and a secondary source. While the primary data was collected from the regular students, the secondary data was gathered from the school and office of the Chief Educational Officer (CEO). A total of 1000 datasets of the year 2006 from five different schools in three different districts of Tamilnadu were collected. The raw data was preprocessed in terms of filling up missing values, transforming values in one form into another and relevant attribute/ variable selection. As a result, we had 772 student records, which were used for CHAID prediction model construction. A set of prediction rules were extracted from CHIAD prediction model and the efficiency of the generated CHIAD prediction model was found. The accuracy of the present model was compared with other model and it has been found to be satisfactory.
1. Introduction
The paper motivates predictive modeling of higher secondary students’ performance because many personal, socio-economic, psychological, and environmental factors influence outcomes. It proposes generating predictive variables, identifying influential variables, constructing a CHAID model, and validating it to assist low achievers.
- Higher secondary performance is influenced by personal, socio-economic, psychological, and environmental variables.
- Accurate performance prediction can identify low-achieving students early for individual assistance by educators.
- The investigation aims to generate a data source of predictive variables and identify variables that strongly influence academic performance.
- It constructs a CHAID prediction model using the identified variables for higher secondary students in the Indian educational system.
- The developed model is intended to be validated for higher secondary students in India.
2. Previous Studies
Previous studies link student performance with diverse demographic, academic, psychological, socio-economic, health, and environmental factors and apply varied predictive methods. Reported predictive accuracy ranges from 38.33% to 93%, motivating a seven-class CHAID model using selected influential variables for Indian higher secondary students.
- Studies applied decision trees, perceptron learning, Bayesian nets, instance-based learning, rule learning, random forests, neural networks, and support vector machines.
- 74% predictive accuracy was reported for a two-class pass/fail dataset in a distance-learning study.
- 38.33% predictive accuracy was reported for a four-class response variable in a university C++ course study.
- 93% predictive accuracy was reported for a decision-tree model on a two-class dataset of secondary students in Portugal.
- Prior research examined student performance using demographic, academic, psychological, socio-economic, health, and environmental variables.
- Reported model accuracies ranged from 38.33 to 93%, varying with datasets, record counts, predictive variables, and response-class values.
- The study therefore constructed a CHAID model with a 7-class response variable using highly influential variables selected through feature selection.
3. Methodology
The study built a higher-secondary student-performance database from questionnaire, school, and CEO-office sources, then prepared it for CHAID modeling through variable definition, preprocessing, and feature selection.
- Data collection: Survey and experimental methods combined primary student data with secondary records from schools and the CEO office.
- Data collection: A four-page close-ended questionnaire collected student details, with variables informed by literature, educators, parents, government officials, and educational experts.
- Variable design: Student attributes covered family, school-board, secondary-grade, body-mass, instructional-medium, community, tutoring, study-group, sports, and higher-secondary-grade variables.
- Variable design: Higher-secondary performance was the response variable divided into seven grades: O, A, B, C, D, E, and F.
- Model construction: CHAID recursively partitions categorical data into nodes that minimize response variation within segments and maximize it between segments until stopping rules are met.
- Model construction: Pearson chi-square-based filtered feature selection identified relevant variables for classification, reducing computation time and potentially enhancing predictive accuracy.
4. Results
The study constructs a CHAID tree for student-performance prediction, extracts classification rules, and evaluates accuracy using validation and classification results.
- Model construction: Features with chi-square values greater than 100 were selected for CHAID model construction using STATISTICA 7.The selected variables were treated as highly influencing predictors.
- Model construction: The CHAID tree used 772 instances, produced 11 terminal nodes, and divided students initially by medium of instruction.The first split separated 433 Tamil-medium students from 339 English-medium students.
- Rules extracted: The extracted rule set predicted higher-secondary grades A, B, and C from combinations of instruction medium, prior achievement, school location, living area, and other variables.The rule set focused on the three classes with the largest numbers of objects.
- Rules extracted: Ten-fold cross-validation was applied during model construction, and the 11 terminal nodes yielded 11 classification rules.The rules were generated by following paths from terminal nodes to the root, with pruning removing nodes containing too few objects.
- Model evaluation: 44.69% was the overall CHAID prediction accuracy, correctly classifying 345 of 772 students and exceeding the earlier Al-Radaideh model's accuracy.The classification matrix compared actual and predicted classifications for seven outcome categories.
5. Conclusions
The CHAID model identified several strong indicators of higher-secondary performance and handled a small, unbalanced dataset effectively. However, limited student samples and geographical coverage prevent generalizing the outcomes across Tamil Nadu.
- Conclusions: Medium of instruction, secondary-education marks, school location, living area, and secondary-education type were the strongest student-performance indicators.The model analyzed interrelations among variables used to predict higher-secondary performance.
- Conclusions: The CHAID model handled a small and unbalanced dataset while achieving better predictive accuracy.The authors suggest that Boosting and Bagging could further improve predictive accuracy.
- Conclusions: Generalization was not possible because the student sample was limited and schools covered only a restricted geographic area of Tamil Nadu.The study reports that a hybrid model was being developed to provide a fuller picture of the real scenario.