Optimising agricultural decision-making is a key pillar of global food security. This work covers two questions, core to the farming community: (i) what are the best crops for your soil and climatic conditions, (ii) how many units should we expect so that we can plan and manage risk. We introduce and validate an integrative two-stage machine learning pipeline. Stage1 A classification model based on soil and climatic variables (nitrogen, phosphorus, potassium, pH, rainfall, temperature and humidity) suggests which crop would be most suitable for the inputs; XGBoost achieves 99.32 % accuracy In Step2, the recommended crop laden as a categorical input to regression model that predicts yield (hg/ha) on farm-level variables (Area, Year, pesticides, temperature and rainfall). Random Forest (R2=0.986, MAE =3846.49 hg/ha). To enhance robustness and practical feasibility, we also report on 5-fold cross-validation, feature-group ablation and deployment-oriented computational metrics.