The aim of this study was to compare the three machine learning validation approach like Holdout, K-fold, and stratified for predicting each soil enzyme activities such as Acid phosphatase, Alkaline Phosphatase, Cellulase, Dehydrogenase, Invertase, N-acetyl-glucosaminidase, Phosphatase, Protease, Urease with physical soil features like sand, silt, clay, depth, and chemical soil properties are available nitrogen, available phosphorus, soil organic carbon, soil organic matter, and other components like PH value, soil fertility level such as low, medium and high. This study used different machine learning algorithms random forest, extra tree, AdaBoost, support vector machine, logistic, ridge, k-nearest, and decision tree algorithm to predict the soil enzyme activity. Compare all the machine learning algorithms and artificial neural networks for calculating better accuracy using classifier algorithm, and also calculate to measure the optimum error using evaluation techniques like means squared error(MSE), root means squared error(RMSE), and mean absolute error(MAE) by regressor algorithm. Suggest the specific crops based on soil properties using a k-means unsupervised machine learning algorithm. In this study, for cellulose, N-acetyl-glucosaminidase enzyme activity by RF, Extra tree, and Adaboost algorithm was better accuracy (99%) using holdout, and K-fold, and stratified validation approach. N-acetyl-glucosaminidase, MSE, RMSE, and MAE measure the optimum error like random forest regressor (RFR) is 0.0094, 0.0712, 0.0155 multiple linear regression (MLR) is 0.005,0.0712, 0.2265 and Decision tree regressor (DTR) is 0.0103,0,0712, 0.0103.