Contextual Outlier Detection on Hotspot Data in Riau Province using k-means Algorithm

Forest fire is one of the environmental problems which has continuously repeated and become a biggest threat to forest resources in Indonesia. One of the forest fire prevention efforts is to determine the spread of hotspots (active fires) clusters. Hotspot data are obtained by remote sensing using satellite that possibly exist the location information containing irregularities (outliers). This research aims to detect contextual outliers on hotspot data in Riau province for the period 2001 to 2009 based on climate context, i.e. rainfall. Contextual outliers were detected using the results of clustering on the daily hotspot frequency attribute and rainfall attribute. The applied method was the technique of clustering using K-means algorithm. The result showed that there were 54 objects detected as contextual outliers, many of them occurred in February, March, June, July, and August. Those objects as contextual outliers were hotspots which have high daily occurrences with high rainfall. The contextual outliers detected have an average of daily occurrences is 65.76 hotspots with an average of rainfall is 37.15 mm.