Human activities are directly affected by weather events. In particular, extreme weather events like forest fires, global warming, drought-causing high air temperatures make human life challenging. The use of reliable and accurate weather prediction models is essential to take precautions against these types of climate events. As a result, creating models that accurately forecast the weather is critical. The successful development of deep learning-based weather prediction models has largely aided by technological advancements. With high success rate, this paper proposes a Robust Feature Selection based Hybrid Weather Prediction (RFS-HWP) model for weather prediction. The input dataset is initially pre-processed with the help of Missing Data Imputation and Z-score Standardization. After that, the feature selection process is accomplished using Botox Optimization Algorithm (BxOA) to find the optimal subset of features. The selected features are then fed into the Hybrid Deep Gated Tobler’s Hiking Neural Network (HDGT-HNN) model, which classifies weather conditions into three classes as temperature, pressure and humidity. The hyper-parameters of HPC-DBCN are optimized using Hiking Optimization Algorithm (HiOA). The entire implementation is carried out on Python platform using publicly available Jena climate dataset, and many types of performance measures are calculated. Also, the usefulness of RFS-HWP model is proven by comparing its performance to state-of-the-art approaches. As a result, the RFS-HWP outperforms by accomplishing overall accuracy of 99.3% and proven to be an applicable model for weather forecasting systems.