The present disclosure relates to a method and a system for authenticating a user. In one embodiment, one or more input and target data samples extracted from a plurality of physiological and movement signals of the user are processed to train one or more regression models. In real time authentication, the input and target data samples are extracted from the plurality of physiological and activity signals and mapped with trained regression models to determine a regression error. Based on the regression error, an appropriate authentication signal is then generated and transmitted to the user. Using dynamically selected multiple input and target data samples for user authentication increases the accuracy of authentication, thereby reducing possibilities of invalid authentication. Further, the power consumed by the sensors and computation load is reduced by dynamically powering up and powering down of the one or more sensors based on their usage during the authentication process.