A Stacking Ensemble Machine Learning Framework for Predicting Academic Performance and Identifying At-Risk Students in Rural Secondary Schools: Evidence from Aurangabad District, Maharashtra
Abstract
Secondary rural schools need to be able to identify at-risk students prior to the conclusion of Class 10 exams. This research project developed and tested a machine-learning approach for predicting three tasks: pass or fail, percentage scores, and the most probable subject of failure. The dataset consisted of approximately 6,000 actual student files collected by the researcher from secondary schools located within the Aurangabad district. Support for collecting the data was provided by a Zilla Parishad office letter asking for the cooperation of all the schools. The researcher retained available hard copy documentation related to this process. Each file contained 23 academic, behavioural, school, and home-related attributes. Fifteen different algorithms were evaluated and the Stacking Classifier was chosen to predict whether or not a student will pass or fail. The decision threshold for the Stacking classifier was adjusted using the F2 score to minimize the number of students who failed that the model missed. For the score prediction task, Elastic Net was chosen. Attendance and hours spent studying per day were each hypothesized as being associated with pass/fail status. Both attendance (p < .001) and study time (p < .001) were significantly different among students classified as having passed versus those who had failed. On test data, the stacking model produced an area under the ROC curve (AUC = .936), along with an 88.9% recall rate. In its subsequent implementation into CORE6, the model was constrained to use only six relatively simple variables that could be easily gathered from schools: ninth grade marks; attendance; study hours; an infrastructure quality score; level of education attained by parents; and availability of a coach. The findings demonstrate that it is possible to develop an early warning system for identifying students at risk of failing in secondary schools in rural areas based upon a limited amount of input information readily available at the school level.