Crime prediction is revolutionizing the public safety by equipping law enforcement with data-driven insights. With the recent development in machine learning and deep learning techniques, recently developed tools analyze crime trends and predict the high-risk areas with greater precision. Methods such as spatio-temporal analysis and hybrid models are essential for improving the accuracy and dependability of predictions. Nonetheless, guaranteeing high-quality and uniform data continues to be a vital concern, since incomplete or biased datasets can distort predictions. Moreover, issues of privacy and fairness demand immediate consideration, while the complexity of these models frequently restricts their practical applications. To address these issues, systems that are transparent and comprehensible are essential. For enhancing the system’s reliability, it is also necessary to integrate socio-economic and environmental data. Additionally, crime prediction systems must also be designed to scale efficiently and process the real-time data, and to prevent the misuse and maintaining public trust, designing the ethical frameworks are essential. This paper provides the current state of crime prediction methods, their applications, and challenges. It also identifies gaps and proposes strategies for developing more reliable and ethical systems.