The palm oil bagworm inflicts substantial economic losses on the palm oil industry in Malaysia annually. At present, integrated pest management remains the primary method for managing pests and diseases. However, early prediction of pest occurrence can help to assist in managing sustainable palm oils more effectively. Understanding the time window as well as the lag associated with environmental changes at the plot level is fundamental to such pest management strategies. However, few studies consider historical environmental data and recognise the biological and ecological time lags in pest population responses in their predictive models. This study aims to develop a predictive model based on spatiotemporal environmental data to support early warning systems for managing palm oil bagworm infestations. Therefore, this study examines the best time lags from one to six cycles before the census lagged variables, ranging from period 1 to 6 and trains the model using a random forest algorithm under 12 different time window configurations. Key driving factors influencing bagworm occurrences, including rainfall, land surface temperature, humidity, and road density, are investigated. Experimental results indicate that the combination of the first three cycles (Lag 1, 2, and 3) of temporal data before the census date yields the best model performance with a recall rate of 0.76 and an AUC value of 0.68. Among these, relative humidity (RH_lag2; two cycles before the census) and surface temperature (LST_lag1; one cycle before the census) emerges as the most influential predictors of bagworm outbreaks. Finally, the visualisation of model predictions enables plot-based targeted preventive measures in high-risk areas, thereby supporting the Sustainable Development Goal (SDG) in palm oil management.