IMPLEMENTATION OF EXACT MACHINE UNLEARNING ALGORITHM IN CREDIT SCORING CASES

• Main Author: Pratama, Ilham
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/85032
• Institution: Institut Teknologi Bandung

Machine unlearning (MUL) is an approach used to remove the influence of training data on a machine learning model. This removal can be necessary for various reasons, such as privacy protection, security enhancement, usability optimization, and improving model accuracy. This approach aims to produce a model whose performance is similar to that of a retrained model (naïve MUL). This study aims to evaluate the performance differences between naïve machine unlearning and using exact machine unlearning algorithms, specifically the SISA and DaRE Forest algorithms, in a credit scoring scenario. The credit scoring models were developed using the random forest algorithm as the base model. The MUL algorithms, SISA and DaRE Forest, were also developed using the random forest base model to remove the influence of k = 50% of the training data. The evaluation was conducted by analyzing the model accuracy, the time required to retrain the model, and the similarity of prediction outcomes from each model. The results show that the DaRE Forest algorithm outperforms the SISA and naïve models. The model produced by DaRE Forest achieved higher accuracy, faster retraining times, and greater similarity in prediction outcomes.