Detection Model for Potential Flooding Areas Using K-Means and Local Outlier Factor (LOF)

Floods are natural disasters that frequently occur in various regions worldwide. This research develops a machine learning model by combining the K-Means algorithm with Local Outlier Factor (LOF) to identify flood-prone areas based on spatial patterns and height anomalies. By utilizing coordinate and elevation data, it is important to conduct initial identification for vacant areas that will undergo large-scale strategic development, enabling accurate preventive measures and land use strategies. Normalization Cross Correlation (NCC) is used to measure the accuracy by comparing the detection results with the original data from Sentinel-1 satellite imagery. The measurement results show the highest NCC score of $93.42 \%$, with a detection contamination rate of $2 \%$ of the total data, indicating that anomalous data is well detected in accordance with the validation data. The results of the NCC accuracy level show that the detection and validation results are sufficient to provide an overview of the potential flood areas that have been detected by the proposed algorithm. By using the proposed model, identification of flood-prone areas can be done accurately, so that it can help in making better decisions in spatial planning and infrastructure development with effective land use. In addition, the findings from identifying areas with a high level of validation in this research can be used to develop flood mitigation regulations related to infrastructure standardization to reduce the impact of flooding due to increasingly massive land changes.