Detection of faults in an automated manner is a very powerful computing technique. In recent years because of technological development, it is now possible to detect the fault based on raw sensor data patterns. Passive Infrared sensor (PIR) is very useful in motion detection (hu-man, animal etc.) in the deployed field. We have proposed a modified scheme based on the Long Short-Term Memory (LSTM) algorithm for detecting faults in the PIR sensor module. Our mechanism is specifically tuned for the PIR sensor, which achieved 87% accuracy for detecting faults using the raw data pattern of the PIR sensor module. Further, we have computed loss concerning mini-batch size and sample size to determine the modified mechanism's accuracy. We expect that future re-searchers may work on complex sensors and try to detect their faults using a modified LSTM algorithm.