Comparison of the Performance of Fuzzy C-Means and K-Medoids in Modeling Forest Fire Occurrence (Case Study: Saravan Forests, Gilan)

Hyrcanian (Caspian) area is one of the most important vegetation areas in Iran, which due to its antiquity, has a high ecosystem value. On the other hand, this ecosystem is involved in multiple fires every year and loses a significant level of vegetation, so the use of scientific methods to predict places with potential fire risk is very important. This can be used for the conservation management of Hyrcanian forests. Many real-world systems are used in terms of pattern recognition, so proper use of machine learning methods is essential in practical applications. However, the use of clustering-based methods is emphasized as an effective method due to its approach in pattern recognition and output discovery. The purpose of this study was to evaluate the ability and compare the performance of Fuzzy C-Means and k-Medoids clustering in modeling forest fire occurrence with emphasis on the performance capabilities of the algorithm. Due to the existence of periodic fires, the mentioned algorithms were used to improve the level of coding in MATLAB software in order to improve studies in the field of forest fire risk prediction. Model input criteria in this study are recorded fire points, distance to agricultural areas, distance to the road, distance to the river, air pressure, solar radiation, slope, aspect, wind speed, forest type and percentage of canopy density. The results obtained from the fire hazard prediction map of both algorithms show their high ability to predict the fire occurrence model. Also, based on the results of the confusion matrix table of the comparison of the two algorithms, the FCM algorithm showed better performance than the k-medoids algorithm in predicting places with potential fire risk. Therefore, the use of FCM algorithm is suggested as one of the effective methods in differential clustering for future studies.