The extraction of significant features is essential for efficient fault detection and remaining useful life prediction of rolling element bearing. The fault diagnosis and prognosis are the major milestones in condition-based maintenance to avoid the catastrophic failure of machines and productivity loss. Data fusion is the predominant technology for extracting significant features by fusing several original features. In this work, seventy-two original features were extracted from bearing vibration signals using various signal processing techniques. The relevant features subset is selected from the extracted features using the Random Forest feature selection method to reduce the complexity and computational time. The selected features are fused by fourteen dimensionality reduction techniques to extract 2D fault features to discriminate the various types of bearing faults and compute a health indicator for the remaining useful life prediction. A comparison is made between the fourteen dimensionality reduction techniques to identify the most efficient technique. The Bhattacharyya distance and Support vector machine are used to verify fault diagnosis accuracy. A new index based on the weighted sum of Monotonicity, Prognosabilty, Trendability, and Signal-to-noise ratio is computed for selecting the suitable prognosis health indicator. The Long short-term memory technique is used to predict the remaining useful life of bearing. Two real-world rolling element bearing data sets are utilized to validate the proposed method.