Clustering and Deep Learning - based Anomaly detection in gas sensor data

As civilization as a whole, the manufacturing sector, and energy consumption keeps expanding, it is vital to keep developing advanced, automated monitoring applications and instruments that will enhance the accuracy of environmental monitoring data as well as the efficiency of the ecological tracking system. Solutions for environmental monitoring have evolved into SEM systems that now include contemporary sensors and the Internet of Things (IoT). IoT technology, facilitated by wireless sensor networks and IoT devices, expedites environmental monitoring by enabling efficient real-time data collection, aggregation, and transmission. However, there is a possibility of outliers called anomalies, which comprise noisy data because of random errors. These anomalies are because of sensor malfunctioning, hardware error, power supply error, ambient environmental change, abnormal weather phenomena, etc. As anomaly detection deals with reducing the errors in data, humans cannot do this on their own, hence automation is essential, which can be done using machine learning algorithms. Current algorithms give better accuracy with univariate data distributions, but in multivariate distributions, they don’t work that well. There aren’t any high-accuracy algorithms for data with complex anomalies which are often observed in the real world, which gives a further scope for detection of anomalies. This work aims to identify anomaly detection algorithms that can provide better accuracy for multivariate distributions. By putting in our knowledge of sensors working better anomaly detection is possible. As a result, it is feasible to more effectively balance the efficiency and precision of anomaly detection in large data sets containing elevated dimensionality problems.