Real-Time Anomaly Detection for Enhanced Vehicle Safety Using Machine Learning Algorithms

In order to be sure of the safety of the vehicle, there is need to continually observe the most critical subsystems and early identify deviant behavior and avoid it before it leads to mechanical failure or an accident. The author presents a real-time anomaly detection system in this paper and is premised on the utilization of the sophisticated machine learning algorithms to enhance car safety and reliability. Multimodal sensor data can be engine temperature, vibration pattern, brake pressure, fuel injection and battery parameters which are established at high sampling rates by Controller Area Network (CAN) bus and On-Board Diagnostics-Version 2 (OBD-II) interfaces. They are processed, extracted features and time segmentation measures in a wholesome manner to come up with good inputs to five machine learning models that include, Random Forest, Support Vector Machine, Long Short-Term Memory (LSTM), Autoencoder and a hybrid CNNLSTM network. It was found that CNNLSTM model has best detection accuracy of 97.41%, F1-score of 96.61% and has minimum inference time that can be implemented within an embedded automotive system. Micro-grained performance measurement across edge devices, including NVIDIA Jetson Nano and Raspberry Pi 4B, is used to guarantee that end-to-end detection latency is reported to be less than 30 ms, which is safe enough. The system provides immediate alerts and takes protective actions in case of any deviation is encountered and, thus, enables one to discover the flaws prior to its calamitous outcomes. The next-generation intelligent vehicle safety system can be achieved through the proposed framework of real-time anomaly detection that has the ability to offer scalability, high-precision and low-latency solutions.