Using Occupant's Data for Electricity Load Prediction based on Machine Learning

This paper uses machine learning techniques to model monthly electricity consumption data of Dubai, the United Arab Emirates, at a consumer level to predict month-ahead prediction. Mainly, statistical-based multivariate approaches are used in conjunction with weather data for modeling electricity consumption. However, in this study, data of occupant's profile is analyzed along with regular consumption data for modeling and capturing the trend of electricity consumption. First, the two data sets are joined to extract the consumer's profile-driven electricity consumption dataset. And then, a ranking-based feature selection using a correlation approach is used, and further, the outlier removal process is performed to obtain the filtered data. Among various machine learning approaches, the Random Forest approach is found to have a better model with the least RMSE and MAPE across a 5-fold cross-validation approach.