An IoT-Driven Federated Learning Method for Rainfall Prediction Employing Attention Convolutional Recurrent Networks and Golden Jackal Optimization Techniques

Abstract

Forecasting heavy rainfall poses a significant challenge for meteorological departments, as it is closely related to human life and the economy, being also a reason for the natural calamities, such as droughts and floods, that people face around the world each year. The precision of rainfall prediction is of enormous significance for countries such as India, whose economy is mainly dependent on agriculture. Rainfall prediction supports evading floods, saving properties, and human lives. There is a need to improve cooperative platforms for weather forecasting on large-scale meteorological data to address the global climate challenge. With the rapid improvement of Artificial Intelligence (AI), Machine Learning (ML) is progressively becoming popular for forecasting rainfall. Federated Learning (FL) is a subset of ML that has become a popular technology for data analysis for Internet of Things (IoT) applications. The innovative growth of IoT-based methods improves smart agriculture, leading to advanced agricultural systems that progressively increase crop yields, reduce irrigation waste, and make it more profitable. This study presents a Method for Rainfall Prediction by employing Attention Convolutional Recurrent Networks and the Golden Jackal Optimization (EMRF-ACRNGJO) approach. The proposed technique initially performs data preprocessing in various stages, such as missing value handling, duplicate removal, categorical to numerical encoding, and scaling. An Attention-based Convolutional Recurrent Neural Network (A-CRNN) is utilized for classification tasks, using the Golden Jackal Optimization (GJO) method for the hyperparameter tuning process. The EMRF-ACRNGJO method was tested on a benchmark dataset, demonstrating an accuracy of 73.30%.