Nowadays, there has been a sharp increase the criminal activities day by day. More and more crimes are being reported but curbing them has become a very difficult activity. With the limited police force, it becomes much more difficult to maintain the law-and-order situation in the country. In order to tackle the crimes effectively, we must be able understand underlying patterns from the existing crime data and predict which crime might occur in a given situation. In this project, we have considered San Francisco crime classification data set for prediction of crimes. This is an open-source data set from Kaggle which contains about three years of crime data with about 39 crime categories. After intensive exploratory data analysis and comparing the performance on various machine learning algorithms, we have created our model using XGBoost Classifier. Since this is a multi-class classification problem, we have tried to predict the probability of occurrence of different types of crime categories in a given situation. Our model has been deployed and saved on a web application and it is ready to use to explore underlying crime patterns and predict the type of crime that might occur in a given situation. Our model has proved to be accurate and outperforming most of the existing models.