Comparative analysis of instance-based learning and metaheuristic algorithms for environmental change and landslide susceptibility modeling

Abstract

Landslides pose significant threats to lives, infrastructure, and the environment, making accurate susceptibility prediction crucial for disaster risk reduction. However, traditional prediction models often struggle with high-dimensional, complex environmental datasets. This study addresses this challenge by comparing four machine learning models-Lazy IBK, Lazy LWL, Random Subspace, and Multilayer Perceptron (MLP)-with four nature-inspired optimization algorithms-Binary Bat Algorithm (BBA), Future Search Algorithm (FSA), Black Hole Algorithm (BHA), and Whale Optimization Algorithm (WOA)-to enhance prediction performance. The models were evaluated for training and testing datasets using metrics like True Positive (TP) Rate, False Positive (FP) Rate, F-Measure, ROC Area, and PRC Area. Results indicate that Black Hole Algorithm (BHA-MLP) and Whale Optimization Algorithm (WOA-MLP) outperformed all other methods, achieving AUC scores of 0.9704 in training and 0.98 in testing for BHA-MLP and 0.974 in training and 0.9787 in testing for WOA-MLP. These algorithms ranked 1st overall with a total score of 7, showing superior classification accuracy and generalization. Among machine learning models, Lazy IBK ranked highest, scoring 28, excelling in TP Rate and FP Rate. The results clearly indicate that combining MLP with nature-inspired optimization algorithms significantly improved landslide prediction accuracy compared to the standalone models. Future directions include further exploring combining deep learning models with metaheuristic optimization to enhance predictive accuracy in complex environmental datasets.