Machine Learning is an emerging technique for building landslide prediction models in the Philippines. Maintenance data from DPWH-BFDEO reveal that landslides have become a perennial problem to the Benguet First Engineering District, covering the municipalities of Bokod, Kabayan, Itogon, La Trinidad, Sablan, and Tuba, from 2015 to 2021. The occurrence of landslides notably coincided with the wet season in the region. Former landslide prediction models using rainfall time series data developed for the region have been previously optimized through the (a) sub-setting of the landslide inventory according to lithological domains, (b) resampling the datasets, and (c) hyperparameter tuning.
This study aims to further improve the previous ML models produced for road landslide prediction in the Benguet First Engineering District (BFED) through feature selection. The methods will include the computation of the Pearson correlation coefficient and Variance inflation factor as a preprocessing technique. After modeling, the Random Forest attribute called feature importance will be used to determine the most significant rainfall event that caused the landslide events in BFED. Moreover, the effect of feature selection will be validated by an appropriate significance test.
This research addresses the UN Sustainability Goal 11 - Sustainable Cities and Communities and Goal, SDG 13 - Climate Action, and SDG 15 - Life on Land.
