Condition monitoring allows predictive maintenance plans as well as minimizes downtime and lowers operating costs. Bearing failures can be identified much earlier than the stage of fracture propagation, thanks to vibration signals utilized for bearing condition monitoring. Condition-based maintenance comprise of (i) fault diagnosis, which is used to determine the health state of machinery from a set of parameters and (ii) fault prognosis, which helps to detect and isolate early developing faults, thereby increasing the remaining useful life of components. Early failure detection allows one to plan preventive maintenance tasks effectively and avert catastrophic breakdowns. This study presents a novel way for diagnosis and classification. In this research, we propose a new deep learning-based method for detecting anomalies in run-to-failure scenarios of bearing datasets. An algorithm is proposed using Python language wherein both the healthy and fault datasets measured using sensors (accelerometers) are trained in order to predict the earliest fault propagation point. This approach expands on how auto-encoders works by encoding and decoding certain parameters of the IMS bearing dataset and the XJTU bearing dataset along with the incorporation of some statistical parameters using Python. Using this approach, one can determine the fault initiation instance; for example IMS Bearing dataset comprising of 984 files shows failure at point 530 and so preventing maintenance activities can be planned to safeguard useful life of the bearing. Last but not least, this methodology is verified through earlier experimental research.