Enhanced Sensor Fault Detection in Aquatic Monitoring using Deep Learning and Machine Learning Algorithms

This paper deals with the problem of probable sensor failure detection using different Machine Learning and Deep Learning algorithms like VAE, AE, CNN, LSTM, OneClass-SVM, and Isolation Forest in Internet of Things applications, particularly in dynamic environments such as aquaculture, where the reliability and consistency of monitoring systems depend on sensor failure detection. The research shows that the CNN algorithm has a better solution with lower errors, such as MSE and MAE, against other methods under an extensive evaluation using absolute sensor data from an aquatic monitoring system. Further, the hybrid model integrating autoencoder and isolation forest shows outstanding potential to detect anomalies, outperforming conventional models. It offers a reliable solution for fault diagnosis in sensors within an IoT-enabled environment and enhances system credibility.