Weighted Local Complexity Invariance for Time Series Classification

doi:10.16182/j.issn1004731x.joss.21-0456

Abstract

Abstract:

Aiming at the misclassification of existing algorithms for long or unevenly distributed time series, the local complexity information is extracted and weighted local complexity-invariant distance (WLCID) is proposed, which includes the local complexity representation model and the weighted global complexity integration model. Sliding window is used to split up time series, and combined with the complexity-invariant distance, the local complexity information can be extracted. As to the class representation model, the integration weights are quantified with the normalized cumulative between-class distance, with the perspective that the subsequence contributes more greatly with larger between-class distance. Compared with other similar algorithms, the proposed method is good at dealing with the data with uneven complexity distribution and can also performs better in most of the test datasets processing. Besides classification tasks, the improvement in the accuracy of clustering tasks also shows its ability to represent the complexity information of the morphological characteristics of time series.

Key words: complexity-invariant distance, local complexity representation, weighted global complexity integration, class representation model, time series classification

CLC Number:

TP391.9

Yitong Li, Xiaotao Liu, Jing Liu, Kai Wu. Weighted Local Complexity Invariance for Time Series Classification[J]. Journal of System Simulation, 2022, 34(10): 2194-2203.

Figures/Tables 12

Fig. 1

Fig. 2

Fig. 3

Table 1

Table 2

Table 3

Table 4

Fig. 4

Fig. 5

Fig. 6

Table 5

Table 6

References 22

1	Wang W, Shi Y, Luo R. Sparse Representation Based Approach to Prediction for Economic Time Series[J]. IEEE Access (S2169-3536), 2019, 7: 20614-20618.
2	Mehdiyev N, Lahann J, Emrich A, et al. Time Series Classification Using Deep Learning for Process Planning: A Case from the Process Industry[J]. Procedia Computer Science(S1877-0509), 2017, 114: 242-249.
3	Catley C, Stratti H, Mcgregor C. Multi-Dimensional Temporal Abstraction and Data Mining of Medical Time Series Data: Trends and Challenges[J]. Conference Proceedings: Annual International Conference of the IEEE Engineering in Medicine and Biology Society Conference (S1558-4615), 2017, 2008: 4322-4325.
4	Gasparrini A, Armstrong B, Kenward M G. Dlnm: Distributed Lag Non-Linear Models[J]. Statistics in Medicine (S1097-0258), 2017, 29(21): 2224-2234.
5	Shen F, Liu J, Wu K. Multivariate Time Series Forecasting Based on Elastic Net and High-Order Fuzzy Cognitive Maps: A Case Study on Human Action Prediction through EEG Signals[J]. IEEE Transactions on Fuzzy Systems (S1941-0034), 2020, 29(8): 2336-2348.
6	刘文法, 王旭, 张建邦. 基于LS-SVM的装备需求时间序列预测[J]. 弹箭与制导学报, 2006, 26(1): 780-783.
	Liu Wenfa, Wang Xu, Zhang Jianbang. Time Series Prediction of Equipment Demand Based on LS-SVM[J]. Journal of Projectiles, Rockets, Missiles and Guidance, 2006, 26(1): 780-783.
7	Bagnall A, Lines J, Bostrom A, et al. The Great Time Series Classification Bake off: A Review and Experimental Evaluation of Recent Algorithmic Advances[J]. Data Mining and Knowledge Discovery (S1384-5810), 2017, 31(3): 606-660.
8	Gong Z, Chen H, Yuan B, et al. Multiobjective Learning in the Model Space for Time Series Classification[J]. IEEE Transactions on Cybernetics (S2168-2275), 2019, 49(3): 918-932.
9	Fawaz H I, Forestier G, Weber J, et al. Deep Learning for Time Series Classification: A Review[J]. Data Mining and Knowledge Discovery (S1384-5810), 2019, 33(4): 917-963.
10	Abanda A, Mori U, Lozano J A. A Review on Distance Based Time Series Classification[J]. Data Mining and Knowledge Discovery (S1384-5810), 2019, 33: 378-412.
11	Górecki Tomasz, Łuczak Maciej. Using Derivatives in Time Series Classification[J]. Communications in Statistics: Simulation and Computation(S0361-0918), 2013, 26(2): 310-331.
12	Jeong Y S, Jeong M K, Omitaomu O A. Weighted Dynamic Time Warping for Time Series Classification[J]. Pattern Recognition (S0031-3203), 2011, 44(9): 2231-2240.
13	Kate, Rohit J. Using Dynamic Time Warping Distances as Features for Improved Time Series Classification[J]. Data Mining and Knowledge Discovery(S1384-5810), 2016, 30(2): 283-312.
14	Marteau, Pierre-François. Time Warp Edit Distance with Stiffness Adjustment for Time Series Matching[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence (S1939-3539), 2008, 31(2): 306-318.
15	Stefan A, Athitsos V, Das G. The Move-Split-Merge Metric for Time Series[J]. IEEE Transactions on Knowledge and Data Engineering(S1041-4347), 2013, 25(6): 1425-1438.
16	Batista G, Keogh E J, Tataw O M, et al. CID: An Efficient Complexity-Invariant Distance for Time Series[J]. Data Mining and Knowledge Discovery (S1384-5810), 2014, 28(3): 634-669.
17	Amato F, Laib M, Guignard F, et al. Analysis of Air Pollution Time Series Using Complexity-Invariant Distance and Information Measures[J]. Physica A: Statistical Mechanics and its Applications(S0378-4371), 2020, 547: 124391.
18	Zhao X, Ji M, Zhang N, et al. Permutation Transition Entropy: Measuring the Dynamical Complexity of Financial Time Series[J]. Chaos Solitons & Fractals (S0960-0779), 2020, 139: 109962.
19	Biswal B, Biswal M, Hasan S, et al. Nonstationary Power Signal Time Series Data Classification Using LVQ Classifier[J]. Applied Soft Computing (S1568-4946), 2014, 18: 158-166.
20	Whelen T, Siqueira P. Time-Series Classification of Sentinel-1 Agricultural Data over North Dakota[J]. Remote Sensing Letters (S2150-704X), 2018, 9(5): 411-420.
21	Calegari G R, Carlino E, Peroni D, et al. Filtering and Windowing Mobile Traffic Time Series for Territorial Land Use Classification[J]. Computer Communications (S0140-3664), 2016, 95(1): 15-28.
22	Dau, Anh H, Bagnall A, et al. The UCR Time Series Archive[J]. IEEE/CAA Journal of Automatica Sinica (S2329-9266), 2019(6): 1293-1305.

数据集	ED	CID	WLCID
Chinatown	94.20	94.49	97.39
BeetleFly	75.00	70.00	80.00
BirdChicken	55.00	65.00	85.00
FaceFour	78.41	80.68	85.23
ECG200	88.00	89.00	89.00
Herring	51.56	54.69	57.81
Sony1	69.55	81.53	82.70
DPOutlineCorrect	71.74	71.38	74.28
Rock	64.00	66.00	70.00
DSizeReduction	93.46	93.46	94.12
Fish	78.29	78.29	82.86
ECGFiveDays	79.67	78.16	89.31
Trace	76.00	86.00	83.00
MPTW	51.30	50.00	51.30
OSULeaf	52.07	56.20	58.26

数据集	DTW	CID	WLCID
Chinatown	95.36	95.07	96.52
BeetleFly	70.00	70.00	75.00
BirdChicken	75.00	75.00	80.00
FaceFour	84.09	87.50	90.91
ECG200	80.00	80.00	89.00
Herring	54.69	54.69	57.81
Sony1	71.21	81.86	81.03
DPOutlineCorrect	72.46	71.01	71.74
Rock	58.00	46.00	70.00
DSizeReduction	96.08	95.75	96.08
Fish	86.29	85.71	76.00
ECGFiveDays	77.47	78.98	84.20
Trace	99.00	99.00	99.00
MPTW	49.35	50.00	52.60
OSULeaf	63.64	66.12	61.57

方法	Chinatown	Sony1	BME
MSM	93.04	59.73	86.00
MSM+CID	93.91	85.86	93.33
MSM+WLCID	97.68	99.33	89.33
TWED	91.59	52.75	78.67
TWED+CID	91.59	89.85	90.67
TWED+WLCID	97.39	99.33	88.00
MD	95.07	68.72	82.00
MD+CID	95.07	81.70	89.33
MD+WLCID	96.52	69.88	63.33

类别	20个时间序列的CF值
1	1.043	1.131	1.246	1.350	1.348
1	1.113	1.096	1.016	1.441	1.010
2	1.236	1.072	1.063	1.265	1.175
2	1.276	1.005	1.191	1.055	1.146

数据集	ED	CID	WLCID
Chinatown	71.01	71.59	72.46
BeetleFly	70.00	60.00	70.00
BirdChicken	60.00	50.00	65.00
ECG200	73.00	75.00	78.00
Herring	59.38	59.38	50.00
Coffee	53.57	53.57	57.14
Sony1	72.05	61.40	75.04
DPOutlineCorrect	61.59	60.14	60.87
Yoga	50.67	50.73	50.63
GunPoint	52.00	53.33	52.67