A Spatio-Temporal Machine Learning Benchmark for Forest-Habitat Disturbance Prediction around Salonga National Park

Forest-habitat disturbance around Salonga National Park exhibits strong spatial clustering and temporal persistence. Machine learning models, particularly Random Forest, achieved high predictive accuracy (ROC-AUC: 0.980, PR-AUC: 0.980) for one-year-ahead disturbance prediction, outperforming simpler baselines like persistence and historical hotspot models. The study leveraged open geospatial data and a grid-cell-year dataset to classify disturbances based on forest loss thresholds. While tree-based models demonstrated superior performance, the findings underscore the reproducibility of machine learning benchmarks in tropical protected areas and their potential for biodiversity management support.

Mboyo, D. N., Konka, D. N. N. Y., Kambale, E. M., Djenga, G. N., Wayikanga, D., Lucungu, P. B., & Kyamakya, K. (2026). A Spatio-Temporal Machine Learning Benchmark for Forest-Habitat Disturbance Prediction around Salonga National Park. Procedia Computer Science, 290, 812–818. https://doi.org/10.1016/j.procs.2026.08.466