Technical Support for Detection and Prediction of Rainfall

This project evaluates the effectiveness of some DM methods for predicting rainfall using India's historical weather data gathered from the IDM database. Including 10 attributes from the collected meteorological data, only 5 features are generally believed to be significant for precipitation forecasting. The following phases of research are used in the collected data: data cleaning, data selection, Data Transformation and classification. Many pre-processing responsibilities are concerned with determining missing values and eliminating clutter. Hence, two responsibilities are carrying out for finding missing value purpose, which is cleaning and normalization. In this work, six data mining classification methods, including MLR, K-NN, SVM, ANN, Random Forest (RF) and decision tree (DT), were investigated. The classification process divides objects into different classes. A classifier is often trained using a training set, where single or multiple experts assign labels to a set of objects. The proposed RF is a novel classification technique used in rainfall data to classify the region differently from normal rainfall or heavy rainfall. The completed accuracy of the planned RF classifier is 96.1%, which is advanced than other classifiers used in the Rainfall database.

I. INTRODUCTION

The complete datasets used in the study can be imported freely from the network collected from the Ministry of Earth Sciences and India Metrological Department (IMD) (http://www.indiaenvironmentportal.org.in). The dataset corresponds to the data's collected from the overall weather knowledge of India [1]. This work considers 360 data to find rainfall prediction and differentiate the heavy rainfall and normal rainfall [2].