Machine Learning Approach for Crime Prediction: A Comparative Analysis

Crime is a global issue that affects countries at all stages of development. It impacts the local economy, quality of life, and societal well-being, leading to various social problems. In Africa, including Ethiopia, crime rates vary widely across urban and rural regions. One of the main challenges faced by law enforcement agencies, such as the Ethiopian Federal Police Commission, is the lack of sophisticated analytical tools for predicting and preventing illegal activities. This limitation often leads to the use of traditional and time-consuming crime analysis methods, which are often ineffective in addressing the rising crime rates. Previous research has used various algorithms to predict crime, but there is still room for improvement in accuracy and practical application. This research focuses on crimes categorized under criminal law, using a dataset acquired from the Ethiopian Federal Police Commission. The aim is to develop an automatic crime-type prediction model by combining machine learning and deep learning techniques. We chose five models, each bringing unique strengths that contribute to accurate crime-type predictions based on the available data. The first step involved preprocessing the dataset by cleaning the data, normalizing numerical features, handling missing values, encoding categorical variables, and selecting features. The performance of the models was assessed by dividing the data into training and testing sets. Next, we built and trained the predictive models. Each model was trained on the training set using selected features, and techniques like cross-validation were employed to tune hyperparameters and prevent overfitting. Model performance was evaluated using accuracy, precision, recall, F1-score, and confusion matrix metrics to gauge how well each model predicts crime types. Among the models, MLP achieved the highest accuracy at 98.64%, proving the most effective for crime type prediction.