RVFR: Random vector forest regression model for integrated & enhanced approach in forest fires predictions

Abstract

The forest fires is one of the most dangerous disasters to the livelihood planet of earth. Human intervention into the field of the destruction of nature is another cause of these forest fires. The ability to heal nature is dampened due to the expansion of human territories into forests causing loss of lands for the forests. While human expansion cannot be halted, so we as human must take a responsibility for the consequences and make sure the climatic changes due to such disasters, should be as low as possible. Halting the forest fires is an impossible task, and the best we can do control the fires and the amount of area burnt. This process can be improved by predicting the forest fires and the amount of area that might get burnt due to the predicted fire and take measures against it such that there would be no fire at all or the very least reduce the amount of area burnt during the fire. Machine learning in predictions of forest fires is heavily researched and mostly used in real-time. Though usage of machine learning algorithms has been in use, the accuracy of these models is of the lower levels. The paper introduces using a model which is a hybrid of its predecessors, taking as many postulates from them as possible and leaving the cons behind. The hybrid model which is a combination of support vector machine and random forest regression model is using the data of forest fires in the period of 2011–2020 in India, and able to give more accurate predictions. The proposed RVFR model also results in achieving an accuracy of 94%, which is greater than all of the predecessors and with a variance of 1.0.