The goal of this project is to create and assess predictive models for the total farm value of crops produced in various provinces in Canada using agricultural data from the past many years. The research used a rich set of agricultural histories that detailed crop production for many years past, with crop type, average farm price, yield per hectare, production, and acres and hectare seeding area. This project aims to find the most effective machine learning model for total farm value to facilitate agricultural decision-making and future economic planning. A total of thirteen regression models were created and compared based on predictive performance. These models include advanced ensemble learning methods (Extra Trees Regressor, Gradient Boosting Regressor, Random Forest Regressor, Bagging Regressor), single learners (DecisionTreeRegressor, SVR, KNeighbors Regressor), linear models (ElasticNet), boosting methods (AdaBoostRegressor), and a neural network (MLP Regressor). R2, Root Mean Square Error (RMSE), and Mean Absolute Error (MAE) were used as evaluation metrics. The Extra Trees Regressor was determined to be the best model, having a R2 score close to 1.0, while also having the lowest RMSE and MAE score. Ensemble models performed better than other models because they learned a more complex, nonlinear relationship found in the dataset. Fewer adequate models in terms of total farm value prediction were MLP Regressor and the ElasticNet models. The study proves that ensemble machine learning approaches deliver effective results for agricultural forecasting and valuation. The study outcomes will enable policymakers together with farmers and agribusinesses to better estimate yields and select crops while allocating resources effectively. The future research agenda should integrate climate and satellite data to develop spatiotemporal models which will enhance predictive capabilities of these tools.