Landslide Susceptibility Prediction System

Abstract The research presents an innovative landslide susceptibility prediction system that harnesses the power of machine learning and a data-driven approach. This system relies on a robust dataset encompassing five crucial parameters: slope, elevation, precipitation, soil type, and rainfall. To optimize predictive accuracy, four diverse machine learning algorithms—Convolutional Neural Network (CNN), Random Forest, Logistic Regression, and Support Vector Machine (SVM)—are employed. Notably, the system stands out by focusing on real-time predictions without the need for a mapping interface. Users input specific location parameters, and the system leverages selected features to provide instantaneous landslide susceptibility predictions, thus enhancing efficiency while ensuring accuracy. The research outcomes contribute a comprehensive solution, integrating advanced machine learning techniques, a streamlined user experience, and a commitment to swift and precise predictions crucial for decision-making in landslide-prone regions. The iterative and data-driven methodology laid out in the research establishes a solid foundation for continuous refinement and adaptation to evolving environmental conditions, thereby ensuring the system's exceptional performance attributes in terms of predictive accuracy, real-time functionality, user efficiency, and long-term adaptability. This approach holds promise for addressing challenges in landslide management by providing a cutting-edge tool that combines accuracy with user-friendly features and adaptability to changing conditions. The impact of these varying accuracies is significant in shaping the practical implications of the system. The high accuracy of the CNN (with accuracy of 97%) makes it particularly suitable for applications where intricate spatial patterns are crucial for landslide susceptibility assessment. The versatility of the Random Forest model(with accuracy of 93%) makes it adept at handling diverse environmental parameters. Logistic Regression(with accuracy of 97%) might make it suitable for quick assessments, while SVM(with accuracy of 90%) is able to handle non-linear relationships adds a valuable dimension to the overall predictive capability.