Comparative Analysis of Different Sampling Method for Predicting Flood in Special Capital Region of Jakarta

Naufaldo Rafif Basrie, Agus Sihabuddin, S.Si., M.Kom., Dr.

2023 | Skripsi | ILMU KOMPUTER

Floods are increasingly becoming a pressing concern in the Special Capital Region of Jakarta and Kerala, with rising frequencies and intensities posing significant risks to lives and critical infrastructure. In response to these escalating challenges, this research delves into the intricacies of flood occurrences in these regions and explores the imperative of accurate flood prediction. To tackle this issue, our research employs a comprehensive approach, utilizing diverse sampling methods and classifiers for flood prediction. Leveraging climate and flood occurrence data spanning five years (2017-2021),

we draw climate data from the Indonesian Meteorological and Geophysics Body (BMKG) for Jakarta, flood occurrence data from the Indonesian National Board for Disaster Management (BNPB) for both regions, and rainfall data for Kerala from Kaggle. We evaluate the effectiveness of three prominent sampling methods: Regular SMOTE, BorderlineSMOTE, and SVM SMOTE, in combination with two powerful classifiers: Support Vector Machine (SVM) and Random Forest. Through rigorous experimentation and analysis, our research underscores the pivotal role of selecting appropriate sampling techniques and classifiers in achieving precise flood occurrence prediction. Our findings not only enhance disaster response planning but also offer the potential to significantly reduce flood impacts in Jakarta and Kerala. This study provides critical insights to enhance overall flood prediction efficiency in these vulnerable regions.

Floods are increasingly becoming a pressing concern in the Special Capital Region of Jakarta and Kerala, with rising frequencies and intensities posing significant risks to lives and critical infrastructure. In response to these escalating challenges, this research delves into the intricacies of flood occurrences in these regions and explores the imperative of accurate flood prediction. To tackle this issue, our research employs a comprehensive approach, utilizing diverse sampling methods and classifiers for flood prediction. Leveraging climate and flood occurrence data spanning five years (2017-2021),

we draw climate data from the Indonesian Meteorological and Geophysics Body (BMKG) for Jakarta, flood occurrence data from the Indonesian National Board for Disaster Management (BNPB) for both regions, and rainfall data for Kerala from Kaggle. We evaluate the effectiveness of three prominent sampling methods: Regular SMOTE, BorderlineSMOTE, and SVM SMOTE, in combination with two powerful classifiers: Support Vector Machine (SVM) and Random Forest. Through rigorous experimentation and analysis, our research underscores the pivotal role of selecting appropriate sampling techniques and classifiers in achieving precise flood occurrence prediction. Our findings not only enhance disaster response planning but also offer the potential to significantly reduce flood impacts in Jakarta and Kerala. This study provides critical insights to enhance overall flood prediction efficiency in these vulnerable regions.

Kata Kunci : Flood Prediction, Climate Data, Support Vector Machine,Random Forest, SMOTE, Random Oversampling, Random Undersampling