Data Mining for Smart Cities: Predicting Electricity Consumption by Classification

Abstract

Data analysis can be applied to power consumption data for predictions that allow for the efficient scheduling and operation of electricity generation. This work focuses on the parameterization and evaluation of predictive algorithms utilizing metered data on predefined time intervals. More specifically, electricity consumption as a total, but also as main usages/spaces breakdown and weather data are used to develop, train and test predictive models. A technical comparison between different classification algorithms and methodologies are provided. Several weather metrics, such as temperature and humidity are exploited, along with explanatory past consuming variables. The target variable is binary and expresses the volume of consumption regarding each individual residence. The analysis is conducted for two different time intervals during a day, and the outcomes showcase the necessity of weather data for predicting residential electrical consumption. The results also indicate that the size of dwellings affects the accuracy of model.