Optimizing Forest Fire Detection Using PSO, Neural Networks, and k-Fold Cross-Validation

Preventing Forest fires by identifying events based on specific criteria using the Neural Network (NN) algorithm. To ensure result stability, test and training data are divided using k-fold cross-validation with variations k-5, k-10, and $k-20$. The $N N$ algorithm performed significantly better after being optimized with Particle Swarm Optimization (PSO), with the average accuracy value going from $97 \%$ to $98.78 \%$ and the AUC value increasing from 0.997 to 0.999. A weight study of PSO reveals variability in values for each attribute, with Day and Month having no significant effect on classification. The results of the Paired Two-sample t-test for Means show that using PSO improves accuracy significantly. The merging of PSO with Neural Networks has a considerable positive influence on forest fire prediction and is suggested for applications requiring high accuracy.