Smart Grids (SGs) have already achieved wide adoption in information sensing, transmission, and processing. SGs are considered as an advanced digital 2-way power flow power system and are capable of self-healing, adaptive, resilient, and sustainable with foresight for prediction. Data mining play an effective role in SGs in which it enable SGs to be transformed from traditional grids to be an intelligent ones. In this paper, many classification methods which will affect performance of networks in the future are discussed. In fact, classification methods are used in SGs to provide accurate predictions such as electrical load prediction. Electrical load forecasting is a vital process for the electrical power system operation and planning. There are many methods used to improve the load forecasting accuracy in which these methods differ in the mathematical formulation and features used. The classical load forecasting techniques have more complex computational operations with low performance when compared to load forecasting methods based on data mining techniques. A review for feature selection and outlier rejection methods is presented as these processes are very important in data preprocessing phase that enable the prediction models to perform their tasks well.