LPG Leakage Risk Predictions from an IoT-Based Detection System Using Machine Learning

Abstract

LPG has the potential to ignite and explode if it spills since it is a flammable gas. Explosions and flames caused by LPG leaks can damage or kill people who are working in the hazardous area. The industrial sector currently lacks efficient warning systems to identify and anticipate leaks, necessitating efficient equipment to identify and monitor gas leaks. This work aims to investigate and identify a device designed to detect and keep track of LPG leaks by utilizing straightforward yet efficient IoT technologies. The experimental and carefully vetted data are applied to build an Artificial Neural Network (ANN) model for predicting the risk of gas leaks. Gas leaks in the workplace are monitored, alerted to, and controlled by using gas detection systems based on IoT technology. The dataset is then submitted to factor analysis for feature selection, which made use of the model's expertise from its examination of the information gathered from the detecting device in the Cloud system. Several measures were utilized to evaluate the model, including Accuracy, Precision, Recall, F1-Score and Area Under the Curve (AUC). The investigation led to clustering the Risk Ranking Number into three levels, which were then utilized in conjunction with the Risk Matrix to evaluate risk. The net processing time was 1.34 minutes, and the forecast accuracy was 96.05%. This study will assist in improving the model that establishes the alarm system's alert level.