ANN-based and DT-based Classification Approaches to Predict the Rainfall Level of the Grid (90°<I>E</I> − 92°<I>E</I>, 23°<I>N</I> − 25°<I>N</I&a

The study area defined by the coordinates (90°<I>E</I> − 92°<I>E</I>, 23°<I>N</I> − 25°<I>N</I>) is a significant region in Bangladesh, where accurate rainfall predictions are crucial for both the local population and policymakers. Understanding rainfall patterns in this area is vital for effective planning and resource management. Data on atmospheric variables, including temperature, rainfall, humidity, sea level pressure, and wind speed were collected from the Bangladesh Meteorological Department for various locations across the study grids for the period of 1964 to 2015. The descriptive statistics revealed that the pattern of the data of climate parameters is not normal. This dataset serves as the foundation for analyzing climate parameters and forecasting rainfall levels within the specified regions of Bangladesh. This study evaluates machine learning techniques, focusing on artificial neural networks (ANN) and classification and regression trees, C5.0, Random Forest, and Gradient Boosting as alternatives to traditional statistical models for predicting atmospheric phenomena. It reveals that conventional models often rely on assumptions unsuitable for chaotic systems like the atmosphere. Among the assessed models ANN, CART, C5.0, Random Forest (RF), and Gradient Boosting Machines (GBM) the ANN demonstrated the highest predictive capabilities for rainfall forecasting in Bangladesh, achieving superior training accuracy and Kappa values while also being recognized as the best overall performer based on ranking metrics.