A Multidimensionality Reduction Approach to Rainfall Prediction

The rainfall has an impact on various fields and industries, including transportation, construction, tourism, health, and wildlife preservation. Accurate rainfall prediction is essential for mitigating the negative impact of rainfall on these sectors. However, previous studies on rainfall prediction have been mainly based on datasets from North America, Europe, Australia, and Central Asia, covering different periods. This study proposes using weather datasets covering the past 5 to 10 years to capture recent patterns in weather data. Additionally, the curse of dimensionality can impact model performance and lead to overfitting. Therefore, this study proposes utilizing dimensionality reduction techniques to ensure that only the significant features are used for rainfall prediction. Multiple Linear Regression (MLR) with dimensionality reduction is applied to improve the accuracy of rainfall prediction. The experimental result shows that UMAP+MLR and t-SNE+MLR have lower MSEs of 57.27 and 56.74 and higher r2 scores of 0.130 and 0.138, respectively. The proposed approach can be valuable in optimizing resource utilization and mitigating the impacts of rainfall on various fields and industries. The source code for our research is available on GitHub repository: https://github.com/Prasanjit-Dey/Dimension_Reduction.