Short-term load forecasting (STLF) remains a major problem for the proper functioning of power systems. Accurate STLF is essential for decision making in power systems like dispatch scheduling of generation capacity, reliability examination, security valuation, and maintenance planning. But the prediction process becomes tedious due to the variability and non-stationary of load sequence data, particularly in the electricity market. Presently, machine learning (ML) models can be employed for load prediction process without considering the intrinsic characteristics of the load series data. With this motivation, this paper designs an effective Emperor Penguin optimization (EPO) based Least Square Support Vector Machines (LS- SVM) for STLF, called EPO-LSSVM model. The EPO-LSSVM model operates on three major phases namely data pre-processing, prediction, and parameter tuning. Firstly, data pre-processing take place to remove the unwanted data and thereby raises the data quality. Secondly, the LS-SVM model is applied for the prediction of electric load. Besides, the weight and bias parameters involved in the LSSVM model are optimally adjusted by the use of EPO algorithm, which is stimulated from the behaviour of emperor penguins. For examining the enhanced results of the EPO-LSSVM model, a set of simulations were perfumed and inspected the outcomes in terms of different measures on UK Smart Meter dataset. The resultant outcomes highlighted the better predictive results of the EPO- LSSVM model over the other compared methods.