Development of a Rainfall Prediction Model using Convolutional Neural Network Architecture
Activities such as farming, road construction, and travelling, largely depend on rainfall. The prediction of rainfall is crucial so one can plan movement, when to plant, and when to carry out construction work to prevent loss of lives and property. Therefore, this research aims to develop a model that can improve the accuracy of rainfall prediction. The goal of this research is to create a Convolutional Neural Network (CNN) that could predict rainfall. Unlike traditional models that take in weather features in a vector form for rainfall prediction, our model takes these features as a 2D array. This approach allows the CNN model to learn spatial features from the 2D feature array, improving its accuracy. The CNN layer is then followed by the fully connected layer, and finally the regression layer. We compared the proposed model against the traditional approach of using the Correlation Coefficient. The proposed model had a score of 0.6404 as against the traditional approach with a score of 0.622. The integration of the CNN layer improved its accuracy. This research, therefore, shows that transforming weather features into a 2D array has the potential to improve rainfall prediction.
Download Paper