Prep_Seqnet:An Integrated ANN and Sequential Deep Learning Framework for Rainfall Prediction

Rainfall prediction is critical in regions prone to flooding, where changes in weather patterns can significantly impact water levels, transportation, environmental health, evacuation strategies, and urban planning initiatives. Accurate forecasts enable both governmental and private entities to make informed decisions, particularly for disaster response and early flood warnings. Artificial Intelligence (AI) has emerged as a leader in generating detailed rainfall predictions using a variety of methods, including Support Vector Machines, Random Forest, Decision Trees, XGBoost, and others. This study employs AI techniques to forecast the next day’s rainfall in areas susceptible to flooding, analyzing daily weather data spanning a decade from multiple high-risk locations. The proposed integrated ANN and Sequential Deep learning framework named "Prep-SeqNet" incorporates a range of meteorological data, including temperature extremes, precipitation, evaporation, sunlight hours, wind speed, humidity, atmospheric pressure, cloud coverage, and bi-daily temperature readings (at 9 AM and 3 PM). The utilized machine learning architecture is a sequential layered model with four levels, enhanced with dropout for improved regularization. Initially, an ANN base model achieved a 77.5% accuracy rate. However, with the integration of a "Reduce Learning Rate on Plateau" function and a tailored accuracy metric, the refined model exhibited a consistent 18% increase in precision, achieving an accuracy of 95.9%, markedly outperforming the base model.