Over the last few decades, residential electrical load classification and identification have been one of the most challenging research in the area of non-intrusive load monitoring (NILM) for home energy management system. The application of NILM technique in the smart grid has gained enormous attention in recent years. Several methods, including information from the given domains into NILM, have been proposed. Recently, among these methods, machine learning techniques are shown to be significantly better based on large-scale data for load monitoring. In this paper, machine learning techniques are utilized for residential load classification on novel cross-correlation based features, which are extracted from the synthetic time series data. We also present a t-distributed stochastic neighbour embedding (t SNE) based dimensionality reduction from the high dimensional feature set so that the classification can be implemented on a general-purpose microcontroller for near real-time monitoring. Our experimental results show that the extracted features are suitable for reliable identification and classification of different and the combination of residential loads.