Taranath N L, Annmary Jojo, Aqsa Mehareen D, Agampreeth H S "GeoPulse: A Low-Cost Machine Learning Rockfall Prediction System"

Paper_id: 10_6

Abstract

Rockfalls are dangerous phenomena that could destroy the roads, buildings, transportation systems, and it could be dangerous to human beings particularly in mountainous and mining areas. This paper gives an example of a rockfall risk prediction system which estimates the likelihood of a rockfall before it happens based on environmental conditions and time conditions. The main parameters that used to prepare a structured dataset are the intensity of rainfall, the angle of slope, ground vibration, temperature and other time-related characteristics that assist in identifying a gradual change in slope stability that gives useful information on conditions under which rockfall may occur. To make the prediction, two supervised machine learning algorithms, including Logistic Regression and Random Forest were applied to predict the relationship between environmental conditions and past rockfall events. The experimental findings were of good performance, with the Random Forest model recording 97% accuracy and a ROC-AUC value of 0.9841, which means that the model has a high capability of discriminating between safe and risky situations. The obtained results of the confusion matrix indicated numerous correct predictions and a small number of risk cases inaccurately detected, which is significant when dealing with aspects of safety. In the analysis, the two factors that were found to be influential in determining the event of a rockfall were the intensity of rainfall and the slope angle. The system categorises the risk as low, medium, and high, which produces early warnings that aid in minimising the damage and enhance safety in an effective manner.