A Robust Statistical Inference-Driven Subspace Ensemble Classifier for Student Dropout detection in Higher Education

Background/Objectives: Student drop-out is the most burning issues in higher education that induce considerable social and economic costs. The main objective of this study is to predict accurately the dropout risks due to the complexity and diversity of student behaviors. Method: This study has proposed a novel Robust Statistical Inference Fuzzified Random Subspace Ensemble Classifier (ROSIF-RASE). The proposed ROSIF-RASE model focusses on improving prediction accuracy while minimizing time consumption. This model includes several key processes, namely data acquisition, preprocessing, feature selection, and classification. In the data acquisition stage, raw student data samples are collected from the dataset with help of IoT devices. In the second stage, a preprocessing step is performed to convert the input data into a usable format. This step involves two main processes namely imputing missing values and eliminating outliers. To address missing data, the model utilizes robust nonparametric estimator, which estimates the missing values based on known data points. For identifying and managing outliers, statistical residual test is applied for effectively removed from the dataset. After preprocessing, Statistical Inference feature re-weighting algorithm is employed for significant feature selection to reduce the student dropout prediction time. Finally, the classification process is performed using the Fuzzified preferential Random subspace ensemble Classifier to improve the stability and accuracy of student dropout prediction. This process helps to enhance the accuracy of dropout prediction. Experimental assessments are conducted using various evaluation metrics, student dropout prediction accuracy, and precision, sensitivity, and student dropout prediction time. Findings: The proposed ROSIF-RASE model shows measurable performance gains, with increases of 6% in accuracy, 4% in precision, 2% in sensitivity, and 3% in F1-score, along with a minimum 10% decrease in student dropout prediction time. Novelty: The proposed model improves the student dropout prediction accuracy with lesser time compared to different existing methods. Keywords: Student dropout prediction, Robust nonparametric estimator, Statistical residual test, IoT devices, Statistical Inference feature re-weighting algorithm, Fuzzified preferential Random subspace ensemble Classifier