Performance Evaluation Of Machine Learning Algorithms For Rainfall Prediction Using Dimensionality Reduction Techniques

In the last few decades, tremendous change is observed in rainfall patterns which are majorly influenced by two major factors 1) climate change and 2) CO2 emission. Erratic rainfall patterns caused catastrophic effects on agriculture and human life in developing countries like Pakistan, where major economic growth is largely dependent on agriculture. The main objective of this study is to evaluate a performance different Machine learning algorithms for forecasting rainfall patterns using dimensionality reduction techniques on climate change indicators. For this purpose rainfall data was collected for Pakistan. Principle component analysis (PCA), Pearson correlation, and Greedy search algorithms were used for feature selection and the evolution of models was performed using Root Mean Square error (RMSE), Root Absolute Error (RAE), and Coefficient of determination metrics. Results show that features obtained using the Pearson correlation produced the least error and Bayesian linear regression performed with the highest accuracy followed by Neural Network regression.