Towards Effective Time Series Classification of Multi-Class Imbalanced Learning

The problem of classification with multiple classes and imbalanced samples poses a new challenge over the problem of binary classification. Methods for handling imbalanced learning are proposed, but most of them are specifically designed for problems of binary classification. Multi-class inequality poses extra problems for time series, e.g. weather scoring, when used. In this work we introduce the new multi-class imbalanced problem handling algorithm with time series data. Our proposed algorithm is designed to deal with both multiclass imbalances and time series classification issues and inspires the Nearest Neighbour Classification Algorithm of the Imbalanced Fuzzy-Rough Ordered Weighted Average. The feasibilities of our suggested algorithm are tested by an empirical evaluation in a telecom application in Ericsson, Sweden, in which weather classifications are based on data from commercial microwave connections. Our proposed algorithm is compared to the Ericsson model, which is a one-dimensional neural convolution network, and three other model deeper learning. The empirical evaluation shows that our proposed weather classification algorithm efficiency is comparable to that of the current solution. The two most powerful research models are our proposed algorithm and the existing solution.