Intelligent Machine Learning-Based Weather Forecasting Using Hybrid CNN–LSTM and Ensemble Learning

The rapid expansion of meteorological datasets and advances in computational intelligence have accelerated the adoption of machine learning (ML) techniques in modern weather forecasting.  The study investigates the performance of various ML models, including support vector regression, decision trees, ensemble methods, and deep learning architectures such as Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM) networks, and transformer-based models, for predicting key meteorological parameters including temperature, rainfall, humidity, and wind speed.

A hybrid forecasting framework is proposed that integrates CNN layers for spatial feature extraction with LSTM units for temporal dependency learning. The methodology also incorporates Recursive Feature Elimination (RFE) for optimal predictor selection, Principal Component Analysis (PCA) for dimensionality reduction, and k-Nearest Neighbors (k-NN) imputation for handling missing values. Additionally, an ensemble stacking strategy combining Random Forest, Gradient Boosting, and Support Vector Machines is employed to enhance prediction robustness.

Experimental results demonstrate that the hybrid CNN–LSTM model outperforms standalone ML approaches, achieving improved forecast accuracy with reduced RMSE and MAE values across all weather variables. The ensemble stacking layer further boosts generalization and stability, especially under extreme climatic variations. The findings highlight the critical role of hybrid modeling, preprocessing, and ensemble integration in developing intelligent forecasting systems suitable for operational meteorological deployment.