Machine Learning Based Detection of Forest Fires Using Sensors Placed in Forests

Background and Objective: Fire forests can occur for several reasons, and the spreading of fire can be the reason for the partial and complete burning of forests in a large area. It can be dangerous if the fire did not control in the first 15–20 min. Early detection is required to prevent the forest from the effect of fire. For this reason, the early detection of fires can be extinguished in a much shorter time and with less damage. There are many difficulties in detecting fires in forest areas because forest spreads over large areas and are far from residential areas. Due to delayed detection, it becomes increasingly difficult to control the fire and sometimes could be the reason for large-scale damage. This study aimed to determine whether there is a fire in the region using data taken from the sensors placed in the forests using machine learning. Material and Method: The project was carried out according to the following steps: In the study, artificial intelligence techniques were used to create several models for fire detection with the help of sensor data. In this project, machine learning classification algorithms were used to detect fires. The models were developed in the Matlab environment and tested with performance evaluation criteria. Results: As a result of the conducted research, the decision tree models utilized yielded an average accuracy value of approximately 99%, with a positive predictive value of 100%. For ensemble models, the average accuracy was around 97%, accompanied by an average positive predictive value of 96%. In the k-Nearest Neighbors (kNN) models, the average accuracy reached approximately 91%, while the average positive predictive value was about 89%. Regarding Support Vector Machines (SVM) models, an average accuracy of around 93% was achieved and an average positive predictive value of 96%. Through feature selection, model performance, and sensitivity were enhanced up to a success rate of 100%. Conclusion: According to the obtained results, it is evaluated that the developed sensors with the information and the artificial intelligence-based system can be used in practice with a high accuracy rate.