Predicting Without Prejudice: A Deep Learning-Based Bias Mitigation Approach to Student Placement Prediction

In Educational Data Mining (EDM), ensuring accurate and fair student placement prediction is essential for fostering equal opportunities and minimizing biases that may disadvantage certain student groups. This study develops a bias mitigation and explainability framework to enhance fairness and transparency in predictive modeling. Recognizing that data bias can skew prediction outcomes, the study explores various bias mitigation techniques, including re-sampling, re-weighting, and adversarial debiasing, to balance the dataset and ensure equitable representation across student groups. Deep Learning (DL) models are deployed on both the original and bias-mitigated datasets to analyze differences in placement predictions. The results reveal significant disparities in prediction outcomes, highlighting that bias mitigation enhances both predictive accuracy and fairness. Additionally, the integration of explainability techniques, such as SHAP (Shapley Additive Explanations) values, provides insights into feature contributions, promoting transparency and trust in AI-driven decisions. This study underscores the critical importance of addressing bias in EDM and advocates for the integration of bias mitigation and explainability methods to ensure fair and equitable placement predictions. By doing so, it contributes to the development of ethical, accountable, and transparent AI systems in education, supporting data-driven, unbiased decision-making in student placement processes.