Weather variability forecasting model through data mining techniques

Abstract

Climate and weather variability are thought-provoking for world communities. In this apprehension, weather variability imposes a broad impact on the economy and the survival of the living entities. In relation to the African continent country Ethiopia, it is desirable to have great attention on the weather variability. The Ethiopian Dodota Woreda region is continuously affected by repeated droughts. It gives a great alarm to investigate and analyze the factors which are major causes of the frequent occurrence of drought. Although the weather scientists and domain experts are overwhelmed with meteorological data but lacking in analyzing and revealing the hidden knowledge or patterns about weather variability. This paper is an effort to design an enhanced predictive model for weather variability forecasting through Data Mining Techniques. The parameters used in this research are temperature, dew point, sunshine, rainfall, wind speed, maximum temperature, minimum temperature, and relative humidity to enhance the accuracy of forecasting. To improve the accuracy, we used the Multilayer-perceptron (MLP), Naive Bayes, and multinomial logistic regression algorithms to design a proposed Predictive Model. The knowledge discovery in database (KDD) process model was used as a framework for the modeling purpose. The research findings revealed that the aforementioned parameters have a strong positive relationship with weather forecasting in meteorology sectors. The MLP model with selected parameters presents an interesting predictive accuracy result i.e. 98.3908% as correctly classified instances. The most performing algorithm, MLP was chosen and used to generate interesting patterns. The domain experts (meteorologists) validated the discovered patterns for the improved accuracy of weather variability forecasting.