Critical Appraisement of Slope Failure Contributing Parameters for Slope Risk Assessment System of Western Sarawak via Multi StatisticalApproaches with Artificial Neural Network

Hazards related to slope failures often causes significant disruptions in a multitude of aspects to the victims. Thus, mitigating its risk is an utmost importance as the consequences are dire. Conventionally, the risks of slope failure are evaluated through a Slope Assessment System for Malaysia. However, the primary limitation of the system is its unsatisfactory level of accuracy. Furthermore, the system has been known to underwent performance degradation when used in areas outside of its data retrieval region such as Western Sarawak that is located at the Western-most tip of Sarawak consisting of Kuching (both landward and seaward), Serian, and Bau. Being home to the state capital city of Kuching, several higher learning institution, and a large-scale industrial park, the region is currently experiencing rapid development. Albeit development is an indicator of a prospering economy, it can increase the risk of exposure to slope failures as settlements being driven out of safe low-lying areas to regions with uncertain levels of slope failure susceptibility. Thus, this study was focused on critically appraising slope failure contributing parameters for Slope Risk Assessment System of Western Sarawak via Multi Statistical approaches with Artificial Neural Network. Unlike the traditional Slope Assessment System, approaching the development via an Artificial Neural Network ensures that the system was driven purely through the training data, making it free from personal biases. An Artificial Neural Network is a Machine Learning approach that is designed to mimic the human brain based on its decision-making process. Two Artificial Neural Networks models were developed in this study for different purposes, where one was used to develop a Landslide Susceptibility Map for the region, and the other was used to develop a new Slope Assessment System specifically for Western Sarawak. The Landslide Susceptibility Map model was developed with the input variables of aspect, curvature, elevation, soil type, lithology type, Land Use and Land Cover, slope angle, rainfall intensity, and Topographic Wetness Index. The Assessment System model on the other hand was developed with the same variable excluding Land Use and Land Cover, and rainfall intensity. This allows the model to be used without having to wait for the dynamic variable’s presence. The evaluation metrics for both models have shown that the development process was a success. The Landslide Susceptibility Model yielded a Root Mean Squared Error of 0.0057 with the hyperparameter of the model being eight neurons in a single hidden layer, a backpropagation learning algorithm, a learning rate of 0.001, and a maximum step of 1E+8. The predictive performance of said model has yielded a recall of 0.9, and a prediction success rate Area Under the Curve score of 0.99. As for the Slope Assessment System model, the same Root Mean Squared Error rate has been achieved with the hyperparameter of four neurons in a single hidden layer, a learning rate of 0.001, a backpropagation learning algorithm, and a maximum steps of 1E+8. The predictive performance of said model yielded a recall of 1.0, and a prediction success rate Area Under the Curve score of 1.0. The Landslide Susceptibility Map that has been developed from its Artificial Neural Network model has shown great correlation with on-site conditions where slope failure points have been identified to occur. However, its primary weakness was spatiotemporal issues arising from the raster files, where it could accurately predict slope failure risks in areas that has undergone major terrain changes for remediation. This issue was solved by implementing the new Slope Assessment System. The new Slope Assessment System was determined to have a recall score of 1, and an Area Under the Curve score of 0.95. Thus, the new system was deemed to have satisfactory levels of accuracy and can be used in Western Sarawak for slope failure susceptibility prediction.