Penerapan Back Propagation Neural Network untuk Identifikasi Bangunan di Wilayah Rawan Longsor

Abstract

Landslides are a natural disaster that frequently occurs in Indonesia, requiring effective prediction methods for risk mitigation. This research aims to investigate the performance and accuracy of the Back Propagation Neural Network (BPNN) in identifying buildings in landslide-prone areas. The dataset used consists of satellite images and building parameters such as building plans, floor plans, foundations, and topographic elements. The data was normalized using the Min-Max Scaler and divided into training (60%), validation (15%), and test sets (25%). The BPNN model was designed with 8 neurons in the input layer, 30 neurons in the hidden layer, and 3 neurons in the output layer, using ReLU and Softmax activation functions. The results show that the model achieved an accuracy of 90%, with the confusion matrix demonstrating accurate classification for most buildings. Out of the total samples, only 1.2% misclassification occurred in the "Less Safe" category. In conclusion, the model achieved an accuracy of 93%, with an average precision of 93.4%, an average recall of 93%, and an F1-Score of 93%. These results indicate that the BPNN model has excellent performance in detecting and predicting the safety level of buildings in landslide-prone areas