Towards an Explainable Multi-Target Regression, for Wear and Friction Prediction for Brake Pad Materials

Abstract

The primary objective of this study is to create an effective multi-target regression model able to predict friction coefficient and wear rate, which are critical parameters for the tribological performance of brake systems. Two models, namely Random Forest (RF) and eXtreme Gradient Boosting (XG), were evaluated using performance metrics such as mean squared error, mean absolute error, and R-squared. In comparing to 1.2, 0.567, 0.59 for RF algorithm, the XG algorithm proves to be the more accurate model with MSE, MAE and R- squared respectively equal to 0.857, 0.4138, 0.756. XG (Extreme Gradient Boosting) outperforms RF (Random Forest) in terms of predictive accuracy in the specified prediction scenario, and the predicted results show good concordance with real values. However, a notable challenge with this model is the lack of interpretability, often referred to as a "black- box." In response to this issue, the study offers a comprehensive explanation, regarding as to how the XG model learns. Shapely Additive explanation model demonstrates that sliding speed is the most influential factor, positively affecting friction coefficient and wear rate of brake pad materials. In summary, the study contributes to the development of a machine learning model, that is accurate and explainable for the prediction of tribological performance in the field of brake pad materials.