Crime forecasting is an emerging technology that aids law enforcement in effectively mitigating and responding to crime, using exploratory data analysis and machine learning techniques. No such system exists in South Africa, a country with ever-increasing crime rates. A crime forecasting system developed for Port Elizabeth, a populous city in the Eastern Cape, is proposed to rectify this issue. Additionally, two further forecasting systems were developed using the well-known Boston and Chicago crime datasets. This was done to compare the performance of each system depending on the input dataset. The k-Nearest Neighbour and categorical Gradient Boosting were the best-performing classifiers with accuracies of 88.4% and 71.16%, respectively. These systems achieved comparable results to the literature and showed that accuracies were within 4% of each other regardless of the dataset. However, the South African dataset was found to be a poor candidate for crime forecasting. Using the South African dataset, the best-performing classifier achieved a low accuracy of 15.58%. The proposed system highlights the urgent need for better database record-keeping for crimes in South Africa.