Data Mining for Evolving Fuzzy Association Rules for Predicting Monsoon Rainfall of India

Abstract

We used a data mining algorithm to evolve fuzzy association rules between theatmospheric indices and the Summer Monsoon Rainfall of All-India and twohomogenous regions (Peninsular and West central). El Nino and Southern Oscillation(ENSO) and Equatorial Indian Ocean Oscillation zonal wind index (EQWFN) indicesare used as the causative variables. Rules extracted are showing a negative relation withENSO index and a positive relation with the EQWIN index. A fuzzy rule basedprediction technique is also implemented on the same indices to predict the summermonsoon rainfall of All-India, Peninsular, and West central regions. Rules are definedusing a training dataset for the period 1958-1999 and validated for the period 2000-2006. The fuzzy outputs of the defined rules are converted into crisp outputs using theweighted counting algorithm. The variability of the summer monsoon rainfall over theyears is well captured by this technique, thus proving to be efficient even when thelinear statistical relation between the indices is weak.