Sustainable agriculture requires an accurate prediction of crop yield, which allows the accurate allocation of resources, reduced impact on the environment, and food security enhancement. The complex and non-linear relationship between climatic factors, soil features, management, and plant growth is not always identified by the traditional statistical analysis techniques. State-of-the-art Machine Learning (ML) and Deep Learning (DL) solutions provide revolutionary fixes by incorporating multi-source datasets (remote sensing imagery, in-season weather, soil profiles, and historical yield records) into predictive frameworks able to model spatial and temporal dynamics. In this proposed study, to promote sustainable agriculture, a machine learning and deep learning hybrid model is proposed, which can help in precisely predicting crop yield. The system combines multi-source information such as satellite-derived remote sensing, in-season weather parameters, soil, and past yield data to model the dynamic spatial and time trends of crop productivity. The approach has integrated XGBoost to structure tabular data modeling, Temporal Fusion Transformer (TFT) to multivariate time-series modeling, and attention in time series, and CNN-LSTM with Attention to spatio-temporal features extraction in multi-date satellite images. To combine model results, the ensemble learning method will be applied to take advantage of the strengths of one model over another in terms of prediction accuracy, generalization, and early-season forecasting. In our results, we find that ensemble-based hybrid ML/DL models that combine tree-based models on structured data with attention-based deep models on imagery significantly and consistently outperform the use of single ML or DL models alone, measuring 10–20% improvement in yield prediction accuracy across multi-region tests. It is a methodological framework that not only ensures high accuracy but also adheres to sustainability objectives since it allows the management of crops to be climate-resilient, the use of inputs to be optimised, and the environmental load to be minimised. The suggested pipeline is scalable, flexible, and provides a clear path towards sustainable agriculture at both data-abundant and resource-constrained conditions.