Energy-Consumption-Prediction-in-Low-Budget-Housing-with-and-without-Time-Series

The aim of this project is to perform data exploration and build models to predict the energy consumption in a low energy building. We have UCI’s appliances energy prediction data set to perform this task. The data set is at 10 minutes for about 4.5 months in which the house temperature and humidity conditions were monitored with a wireless sensor network. Each wireless node transmitted the temperature and humidity conditions for around 3.3 minutes which later was averaged for 10-minute periods. The energy data was logged every 10 minutes with m-bus energy meters. Outside weather data was taken from the nearest airport weather station and merged together with the experimental data sets using the date and time column. Before building the model, we check the null values, transform data types and select variables. After that, we build models for our two version of data i.e. with time series and without time series, based on which we decide the best one based on accuracy and sensitivity since our focus is to check which factors affect the energy consumption. Additionally, we use ROC plot to confirm our decision on model selection. At the end of this project, we get to know the important variables which are impacting the consumption of energy so that the steps could be taken to reduce energy consumption. We modeled our dataset with both qualitative and quantitative method to certain good prediction. It was evident that it is easy to judge qualitative variable than quantitative when it comes to accuracy rate and accurate prediction which was very difficult in quantitative variable prediction. Before building the model, we check the null values, transform data types and select variables using various variable selection and shrinkage techniques such as best subset selection, stepwise selection, ridge and lasso. In addition, principal component analysis has also been used to better visualize the variation present in the dataset. After the variable selection, we build models for our two version of dataset i.e. with time series using methods such as Linear Regression, Random Forest and Gradient Boosting Machines and without time series using Time Series Analysis, depending on which we decide the best one based on accuracy and sensitivity since our focus is to check which factors affect the energy consumption. Among the various models used, Random Forest model gives the best prediction with the minimum square error value. Additionally, we use ROC plot to confirm our decision on model selection. At the end of this project, we get to know what the important variables are impacting the consumption of energy so that the steps could be taken to reduce it. Also, introduction of new variable gave our analysis a support to solve business problem and look at the architectural aspect of building a house and placement of different categories of room which was evidently the most important outcome of our analysis. Station parameters helped us to check our answer in synchronization with inside parameters and to form geo-grading of temperature and humidity which plays an important role is energy consumption. We were also able to justify why certain rooms play important role in energy consumption and how we can minimize energy consumption with simple changes in floor plan.