International Journal For Innovative Engineering and Management Research

Missing data is often an issue in the credit scoring industry, especially in cases where data is MNAR, which means the missing information relates to variables that are not observable.To solve the problem of MNAR data for credit scoring tasks, the proposed method here is an innovative approach known as RMT-NET (Rejection-based Multi-task Network).The three major tasks which RMT-NET simultaneously learns utilizing a multi-task learning architecture are credit score prediction, imputation of missing data, and rejection of untrustworthy data points.The model successfully identifies and removes biased or uncertain data by combining a rejection mechanism, thereby ensuring high accuracy and fairness of credit score forecasts.Employing one common feature representation for each of the tasks through separate branches for every task,.According to experimental results, RMT-NET performs better than traditional techniques in handling missing data and generates credit ratings that are more robust and reliable, especially for datasets containing a significant proportion of MNAR data.The proposed approach is a potential solution for realworld credit scoring applications with imperfect or faulty data because it not only improves the accuracy of predictions but also reduces bias.