A Simulation-Based Analysis of Attendance Patterns and Past Exam Scores in Predicting Student Academic Outcomes

Abstract

This study examines the extent to which student attendance rates and prior exam scores predict final academic performance using a simulation-enhanced multiple linear regression framework. Using a dataset of 708 student records, the analysis incorporated Box-Cox transformation to address assumption violations, Monte Carlo simulation with 1,000 iterations to assess parameter stability, and 10-fold cross-validation to evaluate generalizability. Both attendance and prior exam scores were statistically significant predictors, jointly explaining approximately 46% of the variance in final exam outcomes. Despite relatively high average attendance and prior performance (~78%), the mean final exam score was substantially lower (~58.8%), suggesting potential differences in assessment difficulty or unmeasured contextual factors. Simulation and cross-validation results yielded similar error metrics, indicating that the model is stable and not overly sensitive to sampling variability. These findings highlight the value of combining traditional regression with resampling techniques to produce reliable predictive models and support the use of data-driven monitoring systems for early identification of students at academic risk.