A Comparative Study of Machine Learning Algorithms for Short-term Electrical Load Forecasting

Electrical load forecasting is an important field,
where it can be used to estimate the amount of electricity
needed, the price of electricity, and the number of generators to
be used. Such forecasting can be performed using machine
learning approaches. Therefore, in this paper, we compare four
machine learning algorithms that can be used to predict
electrical load. These algorithms are Support Vector Machine
(SVM), Least Square Support Vector Machine (LSSVM),
Gradient Boosting Machines (GBM), and Random Forest
Regression (RF). The comparison is done on three months
hourly recorded data set that is publicly available from
Pennsylvania Jersey Maryland (PJM) company. Our
contribution is the identification of which algorithm has a
better accuracy and less execution time. This enables electricity
companies concerned with load forecasting to choose an
algorithm from the existing ones as fast as possible without
wasting time searching for an appropriate algorithm. The
results show that RF achieves the best accuracy, and also has
the least execution time.