Title: Predicting Default Payment of Credit Card Clients Using ANN 'Artificial Neural Network' Analysis
Authors: Mohammed N. Jamala ,\ Samy S. Abu-Naser
Volume: 10
Issue: 7
Pages: 86-93
Publication Date: 2026/07/28
Abstract:
This research applies an artificial neural network (ANN) to predict whether a credit card client will default on their payment in the following month, using the 'Default of Credit Card Clients' dataset from the UCI Machine Learning Repository. The dataset contains 30,000 records of Taiwanese credit card holders described by 23 attributes, including credit limit, demographic information, repayment status history, bill statement amounts, and previous payment amounts, with a binary target variable indicating default (1) or non-default (0) in the following month. The network was trained using a feed-forward architecture with a gradient descent optimization algorithm and the normalized squared error as the loss function, complemented by scaling and bounding layers to normalize inputs and constrain outputs to valid probability ranges. The trained network, with an architecture of 23 input nodes, 3 hidden nodes, and 1 output node, achieved a validation accuracy of 80.27% within a 10% error tolerance range, with the payment amount variables (PAY_AMT1, PAY_AMT2, PAY_AMT3) and bill statement amounts emerging as the most influential predictors of default risk.