Data Mining Attribute Selection Approach for Drought Modelling : A Case Study for Greater Horn of Africa

Abstract

The objectives of this paper were to 1) develop an empirical method for selecting relevant attributes for modelling drought and 2) select the most relevant attribute for drought modelling and predictions in the Greater Horn of Africa (GHA). Twenty four attributes from different domain areas were used for this experimental analysis. Two attribute selection algorithms were used for the current study: Principal Component Analysis (PCA) and correlation-based attribute selection (CAS). Using the PCA and CAS algorithms, the 24 attributes were ranked by their merit value. Accordingly, 15 attributes were selected for modelling drought in GHA. The average merit values for the selected attributes ranged from 0.5 to 0.9. The methodology developed here helps to avoid the uncertainty of domain experts' attribute selection challenges, which are unsystematic and dominated by somewhat arbitrary trial. Future research may evaluate the developed methodology using relevant classification techniques and quantify the actual information gain from the developed approach.