DEVELOPMENT OF CLASSIFICATION MODEL TO PREDICT FLOOD-PRONE LOCATIONS USING GEOSPATIAL ARTIFICIAL INTELLIGENCE AND SNI 8197:2015 METHOD (CASE STUDY : GEOSPATIAL INFORMATION AGENCY)

• Main Author: Setya Nugroho, Yudha
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/67211
• Institution: Institut Teknologi Bandung

Abstract— Geospatial information agency as data custodian of flood-prone requires a breakthrough innovation to predict flood-prone locations in real time. Currently, flood-prone data processing is carried out offline through a desktopbased spatial software. The emergence of integrated artificial intelligence technology with geospatial science produces geospatial artificial intelligence. Flood-prone class categories such as low, medium, and high can be linked to classification model with techniques of supervised learning. The aim of research gap to be carried out to develop machine learning classification models using spatial mapping bases. Focus on this research are development classification models in terms of the application of geospatial datasets, feature selection analysis using the SNI 8197: 2015 method with similar other research. At the moment, The algorithms for the classification model are Support Vector Machine(SVM), Random forest (RF), and MLP-ANN based on comparative studies with the best accuracy. DSRM is used as aguideline for the development of geospatial systems models and prototypes. The application of the random forest classification algorithm, geometry and 3 (three) features, namely rainfall, slope, and land cover resulted in a floodprone machine learning model. The result of average accuracy from classification model using parameters of SNI 8197:2015 compared to other research get an increase value of 1,48%. Meanwhile, the results of performance evaluation from classification model using the multiclass confusion matrix has result an increase in accuracy correction with a value of 1.2%. Thus, research contributions can be achieved by increasing the accuracy. The final result of the performance evaluation from classification model has increased accuracy with a total of 2.68%..