Explainable Artificial Intelligence (xAI) for Interpreting Machine Learning Methods and Their Individual Predictions

Machine learning (ML) methods are shown to provide better predictions than simpler linear methods. Still, academics and practitioners in finance tend to use linear methods as they are considered easier to interpret. I show how to make ML methods equally interpretable by using an explainable Artificial Intelligence (xAI) approach based on SHapley Additive exPlanations (SHAP). The approach provides intuitive explanations of ML methods' global behavior, their predictions at the individual level, and the non-linear relationships between predictors and predictions that they learn from the data. I demonstrate the approach empirically by using the ML method Extreme Gradient Boosting (XGBoost) to predict bankruptcy with a comprehensive real-world dataset of 159,655 recent financial statements published by privately held SMEs. I hope this study paves the way for more use of ML to solve problems in finance and other domains. For this purpose, I also share my code