Performance Analysis of Integrated GeoAI Using Multi-Layer Perceptron Based on Hyperparameter to Identify Landslide Prone Areas: A Case Study from West Bandung

One of the key challenges in disaster mitigation is the availability of information for early warning efforts. However, the information that is available is inherently periodic, while the need for continuous information delivery to the public is essential. Conventional methods often yield information with low accuracy. Therefore, in this study, we evaluate data processing to be interpreted into topographical modeling of landslide-prone areas by integrating two disciplines, namely informatics and geodesy, into a unified field of Geographic Artificial Intelligence (GeoAI). This research aims to evaluate the performance of a multi-layer perceptron (MLP) based on hyperparameters to classify landslide risk levels. The highest accuracy was achieved by MLP 2 with a value of 80%, but this model experienced overfitting, causing its validation test to not align with historical landslide data. The most optimal model was MLP 3, with a model accuracy of 70.66% and an overall system accuracy of 70.07%. The model's prediction results will be visualized as a map and presented on a website hosted on a local computer.