Implementasi Algoritma Support Vector Machine (SVM) untuk Deteksi Banjir

– Flooding is the most common disaster in Indonesia in 2022. Based on a report from the National Disaster Management Agency (BNPB), the number of floods equalled 43.1% of the total national disaster events of around 1,524. One of the areas that was repeatedly affected by flooding from 2022 to 2023 was South Sumatra 99 times. To overcome this problem, machine learning technology can be used as a detection and early warning tool against flooding, one of which is the SVM algorithm. In this study, the performance of various SVM kernels was assessed, and the kernel with the best performance was selected for use in the flood detection system. This research categorizes the flood status with three classification labels: "safe," "alert," and "danger." Various SVM algorithms such as linear, polynomial, RBF, and sigmoid are implemented using public datasets to build a flood status prediction model. Then it will be tested with a flood detection device consisting of an Arduino Uno, NodeMCU, and various sensors such as ultrasonic, water discharge, rainfall, and temperature and humidity sensors. Evaluation measures accuracy, precision, recall, F1-Score, ROC curve, and Cross-Validation. The evaluation results showed that the SVM model with a polynomial kernel was selected as the optimal detection model compared to other kernels. This model achieved a training and testing accuracy of 1.0, a training time of only 0.0012 seconds, a prediction time of 0.0002 seconds, and precision, recall, and F1-score of 1.0. In addition, cross-validation also reached 1.0 in classifying flood data. Testing on the tool used 131 test data with an accuracy of 1.0. Classification results and sensor data are presented through an Android application, making flood monitoring easier.