Spatiotemporal analysis and intensity prediction of forest fires using cuckoo search hybrid models

Forest fire forecasting is a critical aspect of environmental conservation and ecological risk management, particularly in ‎biodiversity-sensitive areas like Uttara Kannada, India. In this research, this article suggests a new hybrid modeling ap-‎proach that combines Cuckoo Search Optimization (CSO) with ensemble machine learning techniques, namely Random ‎Forest (RF) and XGBoost (XGB), for forecasting fire intensity levels. Known as CSORF and CS-XGB, the hybrid models ‎were trained and validated against a spatio-temporally dense dataset from 2009 to 2024, with primary environmental, ‎topographic, and anthropogenic predictors. Aside from classification modeling, spatiotemporal analyses such as Kernel ‎Density Estimation (KDE), seasonal fire patterns, and influence studies on features were performed to determine high-risk ‎seasons and areas. CSO was used to automate the hyperparameter tuning process for both classifiers, yielding a significant ‎boost in performance. The CS-XGB model registered the top accuracy of 99.49%, better than CSORF's 98.99%. Feature ‎importance testing confirmed ecological significance, and humidity, temperature, and rainfall were the top-ranked variables. The work adds a scalable and precise prediction model that can assist in early warning systems and forest manage-‎ment practices‎.