Predicting Credit Card Approvals with Machine Learning and Deep Learning Techniques

Credit card approval has become a crucial task for financial institutions due to the increasing number of credit card applications and the associated risk of defaults. Our study focuses on predicting credit card approvals using a combination of ML and DL models, achieving notable accuracy and F1 scores. In this research, we begin by loading the dataset, and categorizing it into binary classes - Yes and No. Preprocess the data including the remove duplicates, null values, data transformation, SMOTE for class balancing, and standard scaling support the dataset’s integrity. We apply PCA for efficient feature extraction and a methodical split between training and testing sets ensures a comprehensive evaluation of model performance. We implemented several ML models, including DT, KNN, RF, SVM , AdaBoost, and XGBoost. These models exhibited robust performance, with DT and AdaBoost achieving an accuracy of 0.98. The models were evaluated using metrics such as F1 score. We also explored DL models, including ANN, RNN, CNN , LSTM, BiLSTM, GRU , and various hybrid models combining CNN with LSTM, BiLSTM, and GRU, LSTM with GRU, and BiLSTM with GRU. The DL models showcased remarkable accuracy, with some achieving an F1 score of 0.99. Our research highlights ML and DL models’ effectiveness in predicting credit card approvals, automating and enhancing financial decisions. High accuracy and F1 scores demonstrate these models’ capability to improve credit risk management.