Evaluation of Performance of Statistical and ANN Approaches for Prediction of Rainfall

Abstract

Abstract-Prediction of rainfall for a river basin is of utmost importance for planning and design of irrigation and drainage systems as also for command area development. Since the distribution of rainfall varies over space and time, it is required to analyze the data covering long periods and recorded at various locations to arrive at reliable information for decision support. Further, such data need to be analyzed in different ways, depending on the issue under consideration. In the present study, Extreme Value Type-1 (EV1) distribution based on statistical approach and Multi Layer Perceptron (MLP) network based on Artificial Neural Network (ANN) is adopted for prediction of rainfall at Fatehabad and Hansi. The performance of the statistical and ANN approaches used in rainfall predication are evaluated by model performance indicators viz., correlation coefficient, model efficiency and mean absolute percentage error. The study shows the MLP is found to be better suited network for prediction of rainfall at Fatehabad whereas EV1 for Hansi. Index Terms-Artificial I. INTRODUCTIONPrediction of rainfall for a river basin is of utmost importance for planning and design of irrigation and drainage systems as also for command area development. Since the distribution of rainfall varies over space and time, it is required to analyze the data covering long periods and recorded at various locations to arrive at reliable information for decision support [1]. Further, such data need to be analyzed in different ways, depending on the issue under consideration. Out of a number of probability distributions, the family of Extreme Value Distributions (EVDs) includes Generalized Extreme Value, Extreme Value Type-1 (EV1), Extreme Value Type-2 and Extreme Value Type-3 is generally used for rainfall prediction. EVDs arise as limiting distributions for the sample of independent, identically distributed random variables, as the sample size increases. In the group of EVDs, EV1 distribution has no shape parameter as when compared to other distributions and this means that there is no change in the shape of Probability Distribution Function (PDF). Moreover, EV1 distribution has the advantage of having only positive values, since the data series of rainfall are always positive; and therefore EV1 distribution is important in statistics. Apart from this, with the development of Artificial Intelligence (AI), a number of various AI methods have been developed for prediction of rainfall. The new developed methods include Artificial Neural Network (ANN), Adaptive Neuro-Fuzzy Inference System, Fuzzy Logic, Support Vector Machine, Neuro-Fuzzy Network and Evolutionary Optimization Algorithm. Out of these methods, ANN could deal with non-linear and complex problems in terms of classification or forecasting. The ANN models can represent a complex nonlinear relationship and extract the dependence between variables through the training process [2][3]. In the present study, statistical and ANN approaches are adopted for prediction of rainfall for the da...