Classification Model for Hotspot Sequences as Indicator for Peatland Fires using Data Mining Approach

Author: Muhammad Murtadha Ramadhan, Imas Sukaesih Sitanggang, Larasati Puji Anzani

Abstract

One action which can be taken to avoid forest and land fires is to predict where forest and land fires are likely to happen. This can be done by predicting the hotspot as one of forest fires indicators. A hotspot that appears in a sequence for 2 – 5 days can be a strong indicator of forest fires. This study aims to develop prediction model for hotspot emergence in peatlands in Sumatra in 2014 and 2015 using data mining approach. The classification algorithms used are C5.0 and Random Forest which are categorized in Decision Tree model C5.0 additionally results rule-based model. Accuracy of the decision tree model and the rule-based model from C5.0 and Random Forest on the dataset of 2014 is 96.8%, 96.0%, and 85.6%, respectively. Accuracy of the decision tree model and the rule-based model from C5.0 and Random Forest on the dataset of 2015 is 97.1%, 96.6%, and 75.6%, respectively. The attributes that appear from the hotspot classification model are peatlands depth and peatlands type. Hotspots in sequence are most predicted to happen on peatland that have characteristics such as type of peatlands hemist, saprists or fibrists, peatland depth is shallow, medium or deep, and can happen in every type of land use that are used for plantation or other purposes. Field verification is required to be conducted in the future in order to evaluate the prediction model

Paper penuh dapat diunduh di GRDS.

Categories: Portofolio