Recent advancements indicate that the field of Early Time-series Classification (ETSC) can be considered a valuable part of Machine Learning. In cases where it is meaningful to obtain early classification results of temporal indexed data, i.e. before the whole series becomes observable, there are proposed methods in the literature that manage to do so. However, experiments have shown that these methods perform differently, according to the shape and the statistics of each dataset. A way for obtaining more robust behaviour is by devising ensemble ETSC techniques, which rely on combinations of various methods and dynamically choose the best performing ones to utilize. This thesis will research on ways of combining individual ETS classifiers into a single ensemble approach, better performing and more stable.