Publication

Abstract : In this research, we have explored the complex field of credit risk prediction using a various approach that integrates machine learning model for classification and model explainability. Accurate risk assessment, model performance evaluation, and open communication of predictions to stakeholders are the main goals. The project begins with descriptive statistics and exploratory data analysis (EDA), which reveals information on the financial and demographic makeup of credit applicants. Afterwards, outlier identification and elimination through Tukey's method and IsolationForest are done separately to guarantee the validity of ensuing analyses. Our predictive modelling is based on the use of many techniques, including random forest, XGBoost, LightGBM, logistic regression and finally ensemble model. Every model is fine-tuned for peak performance, and a comparative analysis provides a standard for forecast accuracy. We use methods such as SHAP (SHapley Additive exPlanations) and LIME (Local Interpretable Model-agnostic Explanations) to improve the interpretability of our models. With the use of these techniques, we can better understand and convey the judgments rendered by intricate models, promoting openness in the evaluation of credit risk.