Accurate energy analyses and forecasts not only impact a nation's energy stability/security and environment but also provide policymakers with a reliable framework for decision-making. The load forecast of buildings and electricity companies for the arrangement of risk/low-cost demand and supply resources that fulfill future government commitments, plans consumer targets, and respond appropriately for stockholders. This study introduces two novels deep supervised machine learning models, including: (i) fit Gaussian Kernel regression model with random feature expansion (RFEM-GKR); and (ii) non-parametric based k-NN (NPK-NNM) models for buildings and the utility companies load demand forecasts with a higher predictive potential, speed, and accuracy. Five-fold cross-validation is used to reduce prediction errors and to improve network generalization. Real-load consumption data from two different locations (utility company and office building) are used to analyze and validate the proposed models. Each location data is further divided into six different feature selection (MFS) states. Each state is composed of various (16, 19, 17, 09, 16, and 13) types of real-time energy consumption and climatic feature variables. The energy consumption behaviors are then analyzed in terms of the feature significance applied with 5 min, 30 min, and 1-h of time-based on short-, and medium-term intervals. Eleven distance metrics used to measure the number of the neighboring object and the number of objective functions of the model network for accuracy. With less computational time, higher precision, and high penetration levels of multiple input feature variables, the method RFEM-GKR is proven superior. Therefore, because of its high accuracy and stability, the proposed model can be a successful tool to predict energy consumption.