The machine learning algorithm's productivity relies on the features selected for the analysis. Feature selection is the important pre-processing step in the data analysis, especially for the data sets which consist of more variables. This step helps in getting better accuracy by eliminating the unwanted variables in the dataset. This research work includes Optimal feature selection for crop prediction, and improving crop yield prediction by considering suitable and important descriptors. Feature shuffling, Single feature performance, and Target mean performance are used for feature selection based on feature importance score. The selected variables are given as input to the Gradient boost regressor, Random forest regressor, SVM, KNeighbor regressor, and Decision tree Regressor for better accuracy. The features selected using the hybrid technique give 90.92% of accuracy for the prediction of yield.