Flood Risk Prediction Dataset: Hybrid Random Forest and Transformer-Based Model in Melaka (Synthetic, 10,000 Records)

This dataset supports the research findings of a Master's project paper entitled "Flood Risk Prediction Using a Hybrid Random Forest and Transformer-Based Model in Melaka." The dataset consists of 10,000 synthetically generated hourly observations from five simulated hydrological monitoring stations (ST01–ST05) in Melaka, Malaysia, covering the period from January 2024 to February 2025. Each record includes the following parameters: rainfall (mm), water level (m), discharge (m³/s), temperature (°C), relative humidity (%), and atmospheric pressure (hPa), along with a flood risk classification label (0 = Low, 1 = Moderate, 2 = High). The data was generated using realistic value ranges drawn from published hydrological literature for the Melaka region, which is uniquely exposed to both the Northeast Monsoon (November–March) and the Southwest Monsoon (May–September). The dataset was used to train and evaluate a hybrid machine learning model combining Random Forest and a Transformer-based architecture for multi-class flood risk classification. Class distribution: Low Risk (32.25%), Moderate Risk (57.98%), High Risk (9.77%). SMOTE was applied during training to address class imbalance. Columns: station_id, timestamp, rainfall_mm, water_level_m, discharge_m3s, temperature_c, humidity_pct, pressure_hpa, flood_risk