Hybrid Approaches to Sensor Fault Detection: Combining Statistical and AI Techniques

Sensor fault detection is essential for ensuring the reliability and efficiency of industrial systems, smart cities, and critical infrastructure. This paper investigates hybrid approaches combining statistical methods with AI techniques, focusing on three machine learning algorithms: Support Vector Machine (SVM), Random Forest, and Decision Tree. Using a dataset from an industrial IoT network, we preprocess the data statistically to identify outliers and extract features, enhancing model performance. The SVM is evaluated for handling high-dimensional data, a Decision Tree for simplicity and interpretability, and a Random Forest for its ensemble learning. Our results show that Random Forest outperforms the others in accuracy, precision, and recall, attributed to its robustness against overfitting and noisy data. This study demonstrates that combining statistical preprocessing with AI techniques, particularly Random Forest, significantly improves sensor fault detection, offering a robust solution for diverse operational settings. Future research will focus on optimizing these models and examining their scalability in larger sensor networks.