An Empirical Study on Enhancing Explainability Through Quantum Machine Learning for Credit Risk Analysis

This study delves into exploratory data analysis (EDA) and the application of both classical and quantum machine learning models for credit score classification, focusing on a multifaceted approach to improve model performance and interpretability. It starts with thorough data preprocessing, which involves encoding categorical variables and standardizing numerical features to prepare the dataset for analysis. Visualization techniques such as pair plots are employed to reveal relationships and correlations between features and the target variable. The classical modelling phase involves training a Support Vector Classifier (SVC), which is evaluated on training and test datasets, and utilizing Kernel Principal Component Analysis (KPCA) for dimensionality reduction to enhance model efficiency. In addition to classical methods, the study explores quantum machine learning approaches by applying quantum feature maps and Quantum Support Vector Classifiers (QSVC), which leverage quantum computing principles to potentially boost classification accuracy. To address model interpretability, Local Interpretable Model-agnostic Explanations (LIME) is used to assess and explain feature importances, providing insights into how different features influence predictions. This comprehensive approach integrates traditional and cutting-edge techniques, aiming to advance the understanding and effectiveness of credit score classification models.