RADAR Performance Prediction for Autonomous Driving Using Machine Learning and Process Data Analysis

This paper employs a multioutput regression technique to predict the performance of RADAR sensors used in autonomous driving. The study aims to enhance prediction accuracy by utilizing real-world RADAR feature process data obtained from the manufacturing process. The analysis is conducted through the core algorithm, the catboost regressor, along with two additional standard models and their corresponding multioutput versions. The research findings demonstrate that the modified NRMSE(MNRMSE) of the multioutput regressor models are up to 0.006 lower than that of individual target models, indicating superior performance of the multioutput approach. The primary algorithm, the multioutput catboost regressor, achieved a MNRMSE of 1.9307. This result validates that multioutput models effectively capture correlations between outputs, reduce redundant data learning, and mitigate overfitting. The findings suggest that multioutput regression models, particularly catboost, can be effectively applied not only to RADAR but also to other industrial process datasets to improve yield. Future research will aim to integrate more comprehensive process data and further refine the model to maximize prediction accuracy. Additionally, applying these models to various industrial fields is expected to provide valuable insights for improving production yield and reducing costs.