top of page
Predictive Modeling in IoT
Energy Use Forecast
Project developed with the purpose of implementing predictive models
to predict energy consumption based on data from a home's wireless IoT sensors.
Three predictive models of machine learning were developed (Machine Learning) supervised based on historical data: lm, Random Forest and SVM. These data were collected by wireless sensors for a period of 10 minutes for approximately 5 months, and the temperature and relative humidity conditions of the esidence were acquired by a network of wireless ZigBee sensors.
Exploratory analyzes were developed with the data provided, identifying periods of higher consumption, environments with less variations, outliers indicating possible appliance problems; resource engineering; identification of the most relevant; correlation; underfitting and overfitting control for the machine learning model to provide greater efficiency; model training and forecasting; error measures between predicted values and those previously observed using the AUC metric; in addition to graphics showing each step of the Data Science process.
Finally, when analyzing all the problems provided by the data, a new algorithm for Machine Learning was implemented: K - Means. With it we separate the predicted data from consumption in 9 different categories which helped to understand periods of higher consumption and consequently offer better insights to the user
bottom of page