CLASSIFICATION AND SPATIO-TEMPORAL ANALYSIS ON IMBALANCED DATASET USING MACHINE LEARNING (CASE STUDY: CRIME DATA IN BANDUNG CITY)

• Main Author: Irfani, Irfan
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/81672
• Institution: Institut Teknologi Bandung

Crime is a serious issue affecting the safety and well-being of society. With the advancement of information technology, the analysis of criminal data using classification and spatial-temporal methods has become a primary focus in crime prevention and management efforts, particularly in major cities like Bandung. This study explores various data classification methods, including Decision Tree, Random Forest, and Discriminant Analysis (Linear and Quadratic), to identify factors influencing crime occurrences. Synthetic Minority Oversampling Technique (SMOTE) and Tomek Link techniques are used to address class imbalance in the crime dataset. The results indicate that the Random Forest model with data sampling techniques and hyperparameter tuning using Grid Search Optimization provides better predictions compared to the Decision Tree and Discriminant Analysis (Linear and Quadratic) models. Additionally, this study applies the Ordinary Kriging (OK) method for spatial-temporal crime analysis, involving three main stages: classifying the distance ????????????, selecting an appropriate semivariogram, and performing Ordinary Kriging (OK). This approach aids in mapping crime location patterns and trends and detecting the spatial concentration of crimes.