Time Series Classification

Time Series Classification

Algorithms

• Feature Engineering
• Nearest-neighbor classification with dynamic time warping (DTW)
  • KNN with DTW
• Kernel methods
  • SVM with GK
• Shapelet-based algorithms
  • Shapelet Transform
  • Learning Shapelet
• Tree-based algorithms
  • Time series forest
  • Time series bag-of-features
  • Proximity forest
• Bag-of-words (dictionary-based) approaches
  • Approaches based on discretizing raw time series
    • Symbolic Aggregation approXimation (SAX)
    • Symbolic Aggregation approXimation in Vector Space Model (SAX-VSM)
  • Methods based on discretizing Fourier coefficients
    • Symbolic Fourier Approximation (SFA)
    • Bag-of-SFA-Symbols (BOSS), BOSSVS, RBOSS, SP-BOSS, Randomized BOSS
    • Temporal Dictionary Ensemble
    • Word Extraction for Time Series Classification (WEASEL), WEASEL+MUSE
• Imaging time series
  • Recurrence plot
  • Gramian angular field
  • Markov transition field
• Deep learning
  • Multilayer perceptron
  • Fully CNN
  • Residual Network
  • Encoder
  • Multi-channel CNN
  • Time CNN
  • InceptionTime
• Random convolutions
  • Random Convolutional Kernel Transform (ROCKET), MiniROCKET, MultiROCKET
• Ensemble models
  • Collective of Transformation-Based Ensembles (COTE), Flat-COTE, HIVE-COTE
  • Time Series Combination of Heterogeneous and Integrated Embedding Forest (TS-CHIEF)

Deep Learning based Algorithms

• Generative Models: unsupervised training step that precedes the learning phase of the classifier, the goal is to find a good representation of time series prior to training a classifier
  • Auto Encoders: stacked denoising auto-encoders (SDAE), generative CNN-based, DBN, RNN auto encoder + classifier
  • Echo State Networks: traditional, kernel learning, meta learning
• Discriminative Models: a classifier (or regressor) that directly learns the mapping between the raw input of a time series (or its hand engineered features) and outputs a probability distribution over the class variables in a dataset
  • Feature Engineering:
    • image transform (Gramian fields, recurrence plots, Markov transition fields),
    • domain specific
  • End-to-end:
    • MLP,
    • CNN
    • FCN
    • Residual Network: Resnet
    • Hybrid

References

• faouziTimeSeriesClassification2022
• ismailfawazDeepLearningTime2019 hfawaz/dl-4-tsc: Deep Learning for Time Series Classification
• hfawaz/ijcnn19ensemble: Deep Neural Network Ensembles for Time Series Classification