Answers are grounded in this paper and cite their sources.
Missing values are common in various applications and pose a significant challenge to Machine Learning (ML), Deep Learning (DL) and statistical model performance. The issue of missing credit risk data severely reduces the effectiveness of existing models due to incomplete data patterns, making imputation more difficult. Therefore, developing accurate imputation approaches for missing data is crucial, especially given the high missing rate, complex data patterns, and arbitrary missing patterns in credit risk assessment datasets. To address this issue, imputation method named Spatio-Convolutional Generative Adversarial Imputation Networks (S-CGAIN) is proposed to predict missing credit data in banks. In this research, the Pima Indian Diabetes dataset is used to evaluate the proposed model, with a preprocessing technique called standardization, employed to scale the features and enhance model efficiency. A Bi-LSTM model is used as a classifier for classifying both the missing and presented data, thereby improving classification accuracy. The S-CGAIN provides the best solution for handling missing data by enhancing model performance. The proposed S-CGAIN method achieves an accuracy of 93.23%, a precision of 99.34%, sensitivity of 94.67%, an F1-score of 92.31%, specificity of 95.56%, and an AUC of 91.45%, outperforming existing methods such as Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN) on the Pima dataset.
No takes yet. Share an insight, caveat, or question.
Bansal et al. (2025) studied this question.