Multivariate LSTM-FCNs for Time Series Classification

14 Jan 2018  ·  Fazle Karim, Somshubra Majumdar, Houshang Darabi, Samuel Harford ·

Over the past decade, multivariate time series classification has received great attention. We propose transforming the existing univariate time series classification models, the Long Short Term Memory Fully Convolutional Network (LSTM-FCN) and Attention LSTM-FCN (ALSTM-FCN), into a multivariate time series classification model by augmenting the fully convolutional block with a squeeze-and-excitation block to further improve accuracy... Our proposed models outperform most state-of-the-art models while requiring minimum preprocessing. The proposed models work efficiently on various complex multivariate time series classification tasks such as activity recognition or action recognition. Furthermore, the proposed models are highly efficient at test time and small enough to deploy on memory constrained systems. read more

PDF Abstract

Datasets


Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Time Series Classification ArabicDigits MALSTM-FCN Accuracy 0.99 # 4
Time Series Classification AUSLAN MALSTM-FCN Accuracy 0.96 # 5
Time Series Classification CharacterTrajectories MALSTM-FCN Accuracy 1 # 1
Time Series Classification CMUsubject16 MALSTM-FCN Accuracy 1 # 1
Time Series Classification DigitShapes MALSTM-FCN Accuracy 1 # 1
Time Series Classification ECG MALSTM-FCN Accuracy 0.86 # 1
Time Series Classification JapaneseVowels MALSTM-FCN Accuracy 0.99 # 1
Time Series Classification KickvsPunch MALSTM-FCN Accuracy 1 # 1
Time Series Classification Libras MALSTM-FCN Accuracy 0.97 # 1
Time Series Classification LP1 MALSTM-FCN Accuracy 0.82 # 1
Time Series Classification LP2 MALSTM-FCN Accuracy 0.77 # 1
Time Series Classification LP3 MALSTM-FCN Accuracy 0.73 # 1
Time Series Classification LP4 MALSTM-FCN Accuracy 0.93 # 1
Time Series Classification LP5 MALSTM-FCN Accuracy 0.67 # 1
Time Series Classification NetFlow MALSTM-FCN Accuracy 0.95 # 2
Time Series Classification pendigits MALSTM-FCN Accuracy 0.97 # 1
Time Series Classification SHAPES MALSTM-FCN Accuracy 1 # 1
Time Series Classification UWave MALSTM-FCN Accuracy 0.98 # 1
Time Series Classification Wafer MALSTM-FCN Accuracy 0.99 # 2
Time Series Classification WalkvsRun MALSTM-FCN Accuracy 1 # 1

Methods