Comparative analysis of feature extraction techniques for imbalanced time-series data

Feature extraction plays a vital role in improvising the performance of machine learning models. In real-world scenario, the learning data is imbalance in nature. In this paper, the comparative investigation is performed over imbalanced time-series data. For this some best traditional approaches and machine learning approach are selected such as Statistical Parameters, Hjorth Parameters, Fractal Dimension, Wavelet-Based methods, and Convolutional Neural Networks (CNN) respectively. The result was evaluated over DEAP dataset and among all feature extraction methods. Hjorth Parameters and Fractal Dimension follow closely to CNN but less than CNN. The highest accuracy was achieved by CNN i.e., 82% and it also outperforms other methods.