A Hybrid Stacked Sparse Autoencoder Based Approach to Optimise the Performance of Meteorological Conditions Predictive Decision System

Accurate weather prediction plays a critical role in environmental monitoring and decision‐making systems, yet remains a challenging task due to the massive volume and high dimensionality of meteorological data. Despite advancements in data collection, many existing methods struggle with efficient feature extraction and computational overhead. Addressing this gap, the present study proposes a novel deep learning‐based framework employing a Stacked Sparse Autoencoder (SSAE) integrated with a Feature Ranking‐Based Feature Selection (FRBFS) strategy to enhance the accuracy and efficiency of weather forecasting. The FRBFS technique involves a two‐stage feature selection mechanism—Filter Feature Ranking followed by Subset Search—to identify the most relevant features, thus optimising the learning process and reducing computational cost. The proposed hybrid model is validated using real‐time and historical meteorological datasets from Weather Underground, supporting both short‐ and long‐term forecast scenarios. Experimental results demonstrate that the SSAE‐FRBFS framework significantly outperforms conventional models such as ANN, CNN, and RNN in terms of accuracy (97.06%), sensitivity (98.15%), and F1‐measure (97.40%), while also reducing processing time and error rates. This study concludes that the proposed approach not only improves predictive performance but also offers a scalable solution for intelligent environmental analytics in the era of Big Data.