Onur Enginar, Ph.D.
Department of Financial Mathematics
September 2023
Supervisor: Ömür Uğur (Institute of Applied Mathematics, Middle East Technical University, Ankara)
Abstract
In this thesis study, we develop a novel deep ensembles based architecture that ensembles transfer learning to reduce the time requirement of deep ensembles without compromising the model’s accuracy. We apply our model to open energy datasets. Moreover, this thesis compares SoTA tabular learning models with deep ensembles and traditional machine learning models and provides a benchmark for the literature. We further develop a feature selection algorithm based on boosted deep ensembles model and compare it with linear feature selection models and tree-based feature selection algorithms.
Keywords: Deep Learning, Energy Forecasting, Tabular Learning, Feature Selection