This dissertation was part of the program of MSc in Data Science in the International Hellenic University. The scope of the study is to use Big Data and Data Mining methods in the prediction of the energy consumption loads of non-commercial buildings in Smart Cities. This task was achieved through 9 cases of deployed models. The load prediction was made through Random Forest, Gradient Boosting, Linear and Extreme Gradient Boosting Regression models in 4 cases on the first stage and 3 cases on the second stage. Hyper-parameter tuning and model optimization through k-Fold Cross Validation and GridSearch CV methods took place in the second case scenarios. The results achieved for load prediction were 85.65% for the first case and 94.38% for the second case. For all the evaluations a dataset of 4.3 million datapoints was utilized, as part of the Building Data Genome Project database. For the building type prediction using load data, the Gradient Boosting and the Random Forest Classification methods were used. The score achieved for this case was 90.83% with some preprocess of the data and without parameter tuning.
Collections
Show Collections
