Evo-TFS: Evolutionary Time-Frequency Domain-Based Synthetic Minority Oversampling Approach to Imbalanced Time Series Classification

TL;DR

Imbalanced time-series classification suffers when minority classes are underrepresented, especially for methods that rely on linear interpolation. Evo-TFS introduces a strongly-typed genetic programming framework that evolves time-series samples by simultaneously optimizing time-domain and frequency-domain characteristics via a sliding-window subseries representation and a joint DTW/DFT fitness. The approach yields higher-quality, more diverse synthetic samples, leading to significant gains over baseline oversampling methods across multiple datasets and classifiers, with demonstrated robustness to parameter settings. While computationally intensive, Evo-TFS offers a principled way to preserve temporal dynamics and spectral structure in synthesized data, boosting practical performance in imbalanced TSC tasks.

Abstract

Time series classification is a fundamental machine learning task with broad real-world applications. Although many deep learning methods have proven effective in learning time-series data for classification, they were originally developed under the assumption of balanced data distributions. Once data distribution is uneven, these methods tend to ignore the minority class that is typically of higher practical significance. Oversampling methods have been designed to address this by generating minority-class samples, but their reliance on linear interpolation often hampers the preservation of temporal dynamics and the generation of diverse samples. Therefore, in this paper, we propose Evo-TFS, a novel evolutionary oversampling method that integrates both time- and frequency-domain characteristics. In Evo-TFS, strongly typed genetic programming is employed to evolve diverse, high-quality time series, guided by a fitness function that incorporates both time-domain and frequency-domain characteristics. Experiments conducted on imbalanced time series datasets demonstrate that Evo-TFS outperforms existing oversampling methods, significantly enhancing the performance of time-domain and frequency-domain classifiers.

Figures (9)

• Figure 1: (a) Visualization of imbalanced time series datasets, in which the sample size of one type is significantly smaller; (b) Generating samples using minority class samples by interpolation methods.
• Figure 2: The flowchart of GP.
• Figure 3: The framework of Evo-TFS.
• Figure 4: (a) The structure of STGP; (b) An example of an STGP program.
• Figure 5: (a) Illustration of calculating the Euclidean distance between time series; (b) Illustration of calculating the DTW distance between time series.