Model Explainability using SHAP Values for LightGBM Predictions

The paper aims at demonstrating the cutting-edge tool for machine learning models explainability leveraging LightGBM modelling. The proposed methodology improves inference performance, training time and significantly reduces the "fitting-to-noise" problem for complex datasets, and additionally, grants more insight into model predictions. Discussed methodology is versatile and finds applications in processing big data sets from different fields of science, ranging from econometry to medical or SHM applications, in which networks of MEMS sensors are used.