Performance analysis of data mining algorithms in the prediction of rain fall

Abstract

Weather being a random phenomenon its prediction has been always a challenge for the meteorologist all over the world. There are number of approaches for predicting this weather based on atmospheric data collected. Rain forecasting is a puzzling, composite, vigorous and mind-boggling task. Rain forecasting pretenses right from the primeval times as a challenging task, because it be influenced by numerous parameters like temperature, wind speed and direction, rainfall, humidity, station level pressure, mean sea-level pressure, dry bulb temperature, dew point temperature and vapour pressure. Various data mining techniques were implemented for rain forecasting. With compared to orthodox methods predicting rainfall rate, the methods that were applying chronological records and data mining technology shows improvement in computing accurate results with more accuracy. Many researchers have done excellent works to construct forecasting models with data mining methods;but in them most just test the predicting accuracy at one particular geographical area. In this paper, we analyzed the performance of k-NN, Random Forest, C5.0 and AdaBoost algorithms on different locations and compared the performance using precision, recall, f-measure and classification accuracy. The daily surface data was collected from India Meteorological Department (IMD), Pune of 3 stations form the period 2005 to 2015. The k-NN algorithm perform better accuracy 98.02 % on Jodhpur dataset with compare to other datasets, the ratio of 90:10 of training and testing records and the value of K is 10. The highest accuracy is 99.270 % of AdaBoost algorithm.