Analysis of unsupervised-learning based approach for the hazard prediction along with the application of the Gaussian method of anomaly detection to reduce the false alarm scenarios in underground coal mines

Safety is a prime concern in the underground mining industry. The wireless sensor networks of hybrid architectures are proposed for monitoring underground mines. Very little work is observed in the field to make an intelligent monitoring system; many are supervised learning-based approaches. Unsupervised and reinforcement learning approaches are significantly fewer in number, and these areas are still unexplored for the underground mine scenarios. This paper proposes the analysis of the K-means clustering algorithm for an intelligent gas hazard monitoring system based on an unsupervised learning method. However, before hazard detection, an anomaly detection algorithm is applied to remove anomalies so there will be fewer chances of false alarms. The F 1 score metric is used to verify the anomaly detection algorithm. Principal component analysis (PCA) is used to visualize the multidimensional gas data. After training the model in various scenarios, it is observed that the model is capable of detecting a hazardous situation with a significant accuracy level.