Application and Analysis of Machine Learning Based Rainfall Prediction

Rainfall predicting is closely related to people's lives, and improving the accuracy of rainfall prediction is of great importance for people’s life and scientific researches. In this paper we attempt to use several machine learning algorithms to analyse and predict rainfall from over 140,000 data recorded at 49 weather stations in Australia. The specific research in this paper is as follows.(1) Explain the limitations of using traditional empirical/physical statistical methods in rainfall predicting through reviewing, and explain the advantages and remaining limitations of current machine learning algorithms applied to rainfall predicting over traditional methods.(2) Different methods are applied to pre-process the data set for both typed and continuous variables to ensure the integrity of the information and to avoid the loss of excessive information in the data.(3)The K-nearest neighbour algorithm, random forest algorithm and support vector machine algorithm are used to model the data for prediction. By comparing the prediction results of each model, summarize the advantages and disadvantages of each model.The prediction results show that the prediction accuracy of the above three algorithms are 82.4464%, 85.6416% and 81.3915% respectively, and the F1-score values are 52.2388%, 61.8736% and 65.1689% respectively. Results of each algorithm are similar in accuracy, and all shows good performance in both accuracy and F1-score.