TS-CHIEF: a scalable and accurate forest algorithm for time series classification

Abstract

Time Series Classification (TSC) has seen enormous progress over the last two decades. HIVE-COTE (Hierarchical Vote Collective of Transformation-based Ensembles) is the current state of the art in terms of classification accuracy. HIVE-COTE recognizes that time series are a specific data type for which the traditional attribute-value representation, used predominantly in machine learning, fails to provide a relevant representation. HIVE-COTE combines multiple types of classifiers: each extracting information about a specific aspect of a time series, be it in the time domain, frequency domain or summarization of intervals within the series. However, HIVE-COTE (and its predecessor, FLAT-COTE) is often infeasible to run on even modest amounts of data. For instance, training HIVE-COTE on a dataset with only 1,500 time series can require 8 days of CPU time. It has polynomial runtime w.r.t training set size, so this problem compounds as data quantity increases. We propose a novel TSC algorithm, TS-CHIEF, which is highly competitive to HIVE-COTE in accuracy, but requires only a fraction of the runtime. TS-CHIEF constructs an ensemble classifier that integrates the most effective embeddings of time series that research has developed in the last decade. It uses tree-structured classifiers to do so efficiently. We assess TS-CHIEF on 85 datasets of the UCR archive, where it achieves state-of-the-art accuracy with scalability and efficiency. We demonstrate that TS-CHIEF can be trained on 130k time series in 2 days, a data quantity that is beyond the reach of any TSC algorithm with comparable accuracy.

Blend-Res2net: Blended Representation Space by Transformation of Residual Mapping with Restrained Learning for Time Series Classification

“…Restrained learning (Section 3.2) enforces the overall network to grow or shrink according to the training dataset characteristics. In Figure 5, we highlight another empirical evidence of superiority of Blend-Res 2 Net over conventional deep architecture for time series [7,14], most recent deep architectures [19,20] and the latest time series specific classification algorithm [21] using average arithmetic and geometric ranks as the statistical measures, where Blend-Res 2 Net is top-ranked followed by TS-Chief and ROCKET.…”

“…Recently, Dynamic Multi-Scale Convolutional Neural Network (DMS-CNN) [19], Rocket (RandOm Convolutional KErnel Transform) [20] have been proposed. TS-Chief [21] got inspired by the strong performance of HIVE-COTE.…”

Feature extraction by grammatical evolution for one-class time series classification

“…In this case tree nodes calculate simple statistics on randomly selected subsequences. Building on this idea, Shifaz et al [50] propose to exploit the strengths of successful TSC algorithms creating three novel splitting criteria. Their ensemble of decision trees achieves competitive performance.…”

“…In the GP literature it is most common to use 50 generations [27], however 40 is also often used [9]. In their review of the GP literature, Poli et al [45] state that the number of generations typically falls in the range [10,50], where the most productive search is usually performed.…”

Incremental and Adaptive Feature Exploration over Time Series Stream

“…Various Shapelet-based approaches have been proposed to optimize both the accuracy [15,16] and the efficiency [17,18] of the classification. Another remarkable attempt [19][20][21] adopting ensemble approaches on several TS representations (e.g., Shapelet-based, similarity-based, interval-based etc.) shows a superior accuracy to one single representation classifiers, where TS features are from different representation domains, and can not be presented in a single form.…”