Rolling bearings are essential components in a wide range of equipment, such as aeroplanes, trains, and wind turbines. Bearing failure has the potential to result in complete system failure, and it accounts for approximately 45 %–50 % of failures in rotating machinery. Hence, it is imperative to establish a thorough and accurate predictive maintenance program that can efficiently foresee and prevent mishaps or malfunctions. The literature has employed a variety of techniques and approaches, from conventional methods to contemporary machine learning (ML) and ML-integrated IoT-based solutions, to categorise bearing faults. This article provides an overview of the most recent research and models used in the classification of bearing faults. The literature summary highlights various significant challenges in current models, such as issues with the classification function, complexities in the neural network structure, unrealistic datasets, dynamic working conditions of rotating machines, noise in the dataset, limited data availability, and imbalanced datasets. In order to tackle the problems, researchers have endeavored to improve and apply different methods, such as convolutional neural networks, deep belief neural networks, and LiNet, among others. Researchers have primarily developed these approaches using datasets from publicly accessible sources. This study also identified research gaps and deficiencies, including limited data availability, data imbalance, and difficulties in data integration. The nascent technologies in the field of problem diagnosis and predictive maintenance are acknowledged as Internet of Things-based ML and vision-based deep learning techniques, which are currently in their initial phases of advancement. Ultimately, the study puts forth several prospective suggestions and recommendations.