Gain Ratio as Attribute Selection Measure in Elegant Decision Tree to Predict Precipitation

Abstract

Prediction of precipitation is a necessary tool in meteorology. To date, it is technologically and scientifically a challenging task for scientists and researchers around the globe. Rainfall is a liquid form of precipitation that depends primarily on humidity, temperature, pressure, wind speed, dew point, and so on. Because rainfall depends on several parameters, its prediction becomes very complex. Approaches such as the back propagation, linear regression, support vector machine, Bayesian networks, and fuzzy logic can be applied, but their rate of prediction is very low, which leads to unpredictable results. This paper aims at improving the prediction of precipitation compared to Supervised Learning in Quest (SLIQ) decision trees, especially when datasets are large. Because SLIQ decision trees take more computational steps to find split points, they consume more time and thus cannot be applied to huge datasets. An elegant decision tree using gain ratio as an attribute selection measure is adopted, which increases the accuracy rate and decreases the computation time. This approach provides an average accuracy of 76.93% with a reduction of 63% in computational time over SLIQ decision trees.