<p>As indoor localization demands accuracy in dynamic environments, Visible Light Communication (VLC) combined with machine learning (ML) has shown great promise. VLC turns ordinary LEDs into smart beacons, guiding devices with a minimum accuracy of centimeters. Through a rigorous system review, the significance of the VLC architecture and its suitability for different localization approaches such as Received Signal Strength (RSS), Time Difference of Arrival (TDoA), Angle of Arrival (AoA), and fingerprinting is focussed on in this current work. In addition, this review is a critical examination of the pivotal role of ML, spanning classical algorithms (e.g., K-Nearest Neighbors, Support Vector Machines, Random Forests) to advanced deep learning models (e.g., CNNs, LSTMs, hybrid networks), discussing their application, advantages, and limitations. Key aspects such as system architecture, positioning techniques, evaluation metrics, and scalability are discussed. In addition, the survey highlights challenges such as environmental variability, device heterogeneity, and the complexities of real-world deployment. Using standardized metrics such as Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and Euclidean error, we synthesize reported performance with real-world deployments. Furthermore, the paper confronts the practical challenges of environmental dynamics, device heterogeneity, and real-world deployment complexities. Finally, we outline promising future research avenues, including sensor fusion, advanced modulation schemes, and the development of standardized datasets. This survey aims to provide researchers and practitioners with a panoramic, state-of-the-art view of ML-aided VLC positioning systems, serving as a reference to guide future innovations in the field.</p>