<![CDATA[The volatility of air quality in Jakarta, which often deteriorates abruptly, demands a proactive early warning mechanism rather than mere real-time monitoring. A major limitation in environmental datasets is the class imbalance, where extreme hazardous conditions are recorded much less frequently than normal conditions, causing them to be overlooked by standard prediction models. This study aims to develop an H+1 (next-day) air quality prediction system by integrating the Random Forest algorithm with the SMOTE (Synthetic Minority Over-sampling Technique) data balancing technique. A Time-Lag feature engineering approach was applied to transform historical data from 2010-2025 into future predictive variables. Experimental results demonstrate that the application of SMOTE successfully improved the model's sensitivity in recognizing 'Unhealthy' categories that were previously difficult to detect. Feature analysis revealed that the accumulation of surface Ozone (O3) and Particulate Matter (PM10) serve as the most dominant indicators triggering air status changes for the following day. This system is intended to serve as a health mitigation reference for the public prior to outdoor activities.]]>
Copyrights © 2026
