Real-Time Multi-Zone Sensor Anomaly Detection Using Few-Shot Machine Learning
Atulya Thakur · 2025
This paper presents a novel approach to real-time sensor anomaly detection in multi-zone environments using few-shot machine learning techniques. The system monitors eight distinct zones (A-H) using streaming sensor data and employs a Stochastic Gradient Descent (SGD) classifier with incremental learning capabilities. The proposed method achieves 91.25% accuracy in detecting sensor anomalies while maintaining low computational overhead through memorylimited historical data retention. Our approach demonstrates the effectiveness of combining statistical threshold-based alerting with predictive machine learning models for industrial sensor monitoring applications.