Enhancement in boiler efficiency via controlled operation could lead to energy savings and reduction in pollutant emission. Activities such as scheduled maintenance could be improved by increasing boiler availability and reducing running costs. However, the time interval between recommended maintenance is varied depending on boilers. The application of fault detection, diagnosis and prognosis (FDDP) in industrial boilers plays an important role in optimizing operation, early-warning of faults, and identification of root causes. This review discusses the application of machine learning (ML)based algorithms (knowledge-driven and data-driven) for FDDP, thus allowing the identification of fit-for-purpose techniques for specific applications leading to improved efficiency, operability, and safety.