Feature Optimization for Rice Field Discrimination Using Sentinel-2A Imagery: A Case Study in the Adi Soemarmo Airport Area and its Surroundings

Indonesia is known as an agrarian country with a very large area of rice fields. However, the high dependence on rice imports indicates an imbalance between land potential and domestic production. One of the main causes is the conversion of rice fields into built-up areas, especially in urban fringe areas such as Colomadu (Karanganyar) and Ngemplak (Boyolali), which are clear examples of urban expansion. This study aims to map rice fields in the area around Adi Soemarmo Airport, which has experienced significant land conversion, using Sentinel-2A imagery and the Random Forest algorithm. This algorithm was chosen due to its ability to handle complex data, automatically evaluate feature importance, and resist overfitting. A total of 16 features were used, consisting of 4 spectral features, 4 vegetation indices, and 8 texture features (variance, entropy, homogeneity, etc.). Feature contribution evaluation was conducted to form four feature configurations tested using the Out-of-Bag (OOB) method, with accuracy evaluated through Overall Accuracy (OA) and Kappa statistics. The results showed a gradual increase in accuracy, with the fourth configuration (features with the highest contribution) achieving the highest OA of 71.69% and a Kappa of 0.6562. These findings confirm that optimal feature selection greatly influences classification accuracy and supports more effective remote sensing-based agricultural mapping.