A SAFETY-ORIENTED DESIGN FRAMEWORK FOR INTELLIGENT ADAS–HMI SYSTEMS UNDER CYBER-PHYSICAL THREATS AND DRIVER DISTRACTION

Abstract

Advanced Driver Assistance Systems (ADAS) integrated with Human–Machine Interfaces (HMI) operate as safety-critical software components within increasingly connected and cyber-physical vehicle ecosystems. While significant progress has been made in driver behavior classification, limited research formally links behavioral prediction to measurable software quality attributes such as reliability, availability, robustness, and security. This study proposes a safety-oriented ADAS–HMI software framework that integrates hybrid machine learning models with formal software quality modeling under cyber-physical threat conditions. Using time-series inertial measurement unit (IMU) data from accelerometers and gyroscopes, engineered features including acceleration magnitude, jerk, rolling variance, and frequency-domain representations via Fast Fourier Transform, are extracted to classify multiclass driving behaviors. A hybrid modeling architecture combining Random Forest, XGBoost, and Long Short-Term Memory networks is employed to enhance generalization and temporal sensitivity. Beyond classification accuracy, predictive performance is translated into reliability using the exponential model , with failure rate approximated by misclassification probability. Safety risk is quantified as , and anomaly detection mechanisms based on statistical thresholds enable fail-safe activation under uncertain or corrupted sensor conditions. Experimental results demonstrate that the Random Forest model achieves 0.849 accuracy and the lowest estimated failure rate, leading to measurable reductions in safety risk and improved robustness under noise perturbation. The proposed framework advances software quality engineering in intelligent transportation systems by integrating behavioral analytics with formal reliability and risk assessment for dependable ADAS–HMI deployment.