Crime prediction is critical in improving police strategies and implementing measures for crime prevention and control. In recent years, machine learning has emerged as a critical way to predictive analytics in this domain. However, few studies have thoroughly compared various machine-learning algorithms for crime prediction. This study investigates the predicting capacities of various machine learning and ensemble approaches using historical public property crime data from a large city in India. Five ensemble models, Random Forest, AdaBoost, CatBoost, Gradient Boosting Machine (GBM) and eXtreme Gradient Boosting (XGBoost) and Four machine learning models, K-Nearest Neighbor (KNN), Support Vector Machine (SVM), Naïve Bayes and Decision Trees are used for crime predictive analysis in this study. The XGBoost model outperformed the other models tested, based primarily on historical crime data. XGBoost being an ensemble approach combines multiple weak classifiers to create an effective classifier. Every weak learner concentrates on the faults made by the preceding ones, enabling the model to refine its predictions and fix errors repeatedly. When compared with other models used in the study, this resulted in higher accuracy.