Studying the Factors Affecting the Risk of Forest Fire Occurrence and Applying Machine Learning Techniques for Detection

Abstract

One of the major factors contributing to environmental dangers, which have a variety of detrimental effects on human existence, is forest fires. Therefore, controlling such a situation and preserving lives requires early forecast, quick identification, and swift actions. In order to find the greatest predictor for spotting forest fires, 517 distinct entries were chosen at various points in time for the Montesano Natural Park (MNP) in Portugal. Firstly, we have to check which factors are causing fire more rapidly the correlation analysis was done between the features. Secondly, Long Short Term Memory (LSTM), Auto Regressive Integrated Moving Average (ARIMA) and Support Vector Regression (SVR) technique was applied on this data set to give the ideal and optimal results in predicting the forest fire. In final stage, Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) were used for predictor’s performance and evaluation. Furthermore, we will compare our proposed model results in last chapter with other artificial and machine learning techniques applied on the same dataset found in the previous studies. The results show that the LSTM approach outperformed the ARIMA and SVR predictors in terms of effectiveness and efficiency. The outcomes also demonstrate that low estimation error as compared to other predictors, the LSTM algorithm offers more accurate forecasts. In comparison to previous methods, the findings and results by LSTM increases prediction accuracy and is acceptable for forest fire prediction.