Abstract:
Accurate weather forecasting is vital for resource planning, disaster management, and informed
decision-making. This study leverages deep learning techniques, including Long Short-Term
Memory (LSTM), Gated Recurrent Units (GRU), and 1D Convolutional Neural Networks (1D CNN),
to predict weather trends from sequential data. Extensive preprocessing was conducted to
ensure data integrity and compatibility with the models. Among the evaluated models, the 1D
CNN and LSTM demonstrated superior performance, achieving regression accuracies of 83.09%
and 83.44%, respectively. The GRU model also performed reliably, with a regression accuracy of
81.72%. A Hybrid CNN-LSTM model was tested but exhibited significantly lower performance,
highlighting the robustness of standalone LSTM and CNN models for weather prediction tasks.
Future work will focus on incorporating larger datasets, enhancing interpretability through
explainable AI, and exploring more advanced architectures. This research underscores the
effectiveness of deep learning models for time-series forecasting and provides valuable insights
into their comparative performance.
