In the attempt towards sustainable agricultural practices, a complete ML framework is introduced in this work to maximize the crop yield, make accurate fertilizer recommendations, and reduce agricultural waste. The research uses several datasets such as crop recommendation datasets, crop yield datasets, and soil nutrient profiles, which can enable the accurate prediction and decision-making process. In order to overcome classic imbalance in the datasets, the SMOTE is applied that will make the training of the models more robust. For prediction of crop and fertilizer, several machine learning algorithms are used, which are DT, MLP-ANN, Bagging Classifier, and Voting Classifier. Agricultural waste detection is accomplished with advanced object detection models YOLOv5, YOLOv7, and YOLOv8 to detect the inefficiencies in production of crops. Experimental results show that Voting Classifier can give better performance for crop and fertilizer recommendation, and accuracy is 100% for it, while YOLOv8 performs better than other YOLOs for waste detection, and the precision is 50.7%. The framework accounts implementation is implemented using the Blade web framework fulfilled providing an interactive platform, where user can input data, get farm crop and fertilizer recommendation data, see the agricultural waste detected and visualize to get agricultural waste, using this framework it combines advanced ML techniques with human interface simple implementation to get practical use for sustainable agricultural decision making process.