In the context of the smart grid, a number of attack scenarios are framed as statistical learning problems using batch- or real-time data collection. Machine learning techniques are used in this methodology. Assist in determining whether a metric has been manipulated. The proposed method includes an attack detector that could possibly use previous system information to succeed due to the fragmentary nature of the assignment. We combine decision- and showcase-fusion, common batch- and online-learning approaches, and supervised and semi supervised learning to model the attack detection problem. In order to uncover unobservable attacks by using statistical learning techniques.