Page 255 - 《软件学报》2020年第10期

张伟 等:一种时间序列鉴别性特征字典构建算法 3231

词生成算法在特征生成上的有效性.
从图 7 中和具体的“高/平/低”对比结果统计可以看出,VLWEA 与模型 ED1NN(60/1/4)和 DTW1NN (53/3/9)
相比显著更好,分类效果具有绝对优势.
图 8 和对比结果统计说明,VLWEA 和 2 个不使用集成分类算法的基于 Shapelet 的分类模型 LS(49/6/10)、
FS(60/3/2)对比同样具有显著优势.
图 9 和具体的精度“高/平/低”对比结果统计说明,VLWEA 和 3 个集成分类模型 COTE(27/7/31)、EE
(45/3/17)、ST(40/6/19)相比,比 EE 和 ST 更好,但比 COTE 略差.
Tabel 4 Accuracies of the feature dictionaries built under eight different conditions (%)
表 4 8 种不同条件下建立的特征字典对应的模型分类精度(%)
特征评价指标 tf-idf tf-idf tf-idf Chi tf-idf Chi Chi Chi
阈值 Dynamic 0.3 0.3 3 0 3 2 2
单词长度 Variable Variable Variable Variable Variable Fixed Fixed Fixed
n 0 语法模型 Unigram Unigram Bigrams Unigram Unigram Unigram Unigram Bigrams
Dataset TDVU T0.3VU T0.3VB C3VU T0VU C3FU C2FU C2FB
Adiac 80.6 80.3 80.9 81.7 81.2 83.3 82.5 83.9
ArrowHead 86.7 86.9 86.8 85.1 87.7 74.3 86.3 85.7
Beef 82.2 75.3 83.3 76.7 80 73.3 73.3 81.3
BeetleFly 95 90 95 95 95 95 95 95
BirdChicken 90 90 90 90 90 85 90 85
Car 86.7 86.7 83.3 85 88.3 85 88.3 85
CBF 99.8 99.7 99.6 99.6 99.6 96.8 98.5 99.1
ChlorineC 75 76.1 75.4 74.6 75 73.5 74.2 75.3
Coffee 100 100 100 100 100 96.4 100 100
Computers 70.4 70.5 70.3 69.1 71.8 64 62.7 66.4
CricketX 76 72.6 78.8 71.9 73.5 75.6 76.4 77.1
CricketY 78.2 79 77.2 79.2 81.3 77 76.7 79
CricketZ 77.3 79.2 78.3 77.1 79.5 78 78.1 79.2
DiatomSR 90.6 91.4 89.7 91 90.8 94.4 93.5 94.1
DistalPOAG 75.5 76 76 77.3 75.2 79.1 78.4 77
DistalPOC 78.7 76.1 75.4 77.6 78.5 76.1 75.2 78.3
DistalPTW 67.6 68.3 67.2 67 67.6 62.6 67.6 69.8
Earthquakes 75 74.8 74.8 74.8 74 74.1 74.8 74.1
ECG200 86.7 85 85.8 85.6 84.4 85 85 84
ECG5000 94 94.2 94.2 94.2 93.9 94.5 94.4 95
ECGFiveDays 100 99.9 99.9 100 99.9 99.9 99.9 100
FaceAll 78.1 78.8 79.1 78.9 79.6 79.2 77.4 77.2
FaceFour 98.9 100 98.9 100 99.3 100 100 100
FacesUCR 94.4 94.1 93.3 93.9 93.9 94.7 94.9 94.7
Fish 97.7 96.6 96.6 96.8 97.5 94.9 96.4 96.6
GunPoint 100 100 100 100 99.3 100 98.7 100
Ham 65.4 64.8 62.9 65 67 62.9 63.8 64.8
Herring 63 64.7 61.9 59.4 63.8 67.2 68.7 60.9
InsectWS 62.7 63.1 63.9 63.6 65 61.7 61.9 64.1
ItalyPD 95.9 95.1 94.9 94.1 95.6 94.8 94.4 94.9
LargeKA 67.8 67.8 68.3 70.6 68.5 60.4 58.5 62.9
Lighting2 67.2 62 63.9 54.1 69.2 52.5 63.9 55.7
Lighting7 71.7 70.2 65.8 69 75.1 76.7 75.3 74
Meat 91.7 91 90 91.7 88.3 90 90 90
MedicalImages 75.5 75.1 74.3 74.7 74.9 70 71.1 75.8
MiddlePOAG 58.8 59.6 58.9 57.9 60.5 54.5 53.2 56.5
MiddlePOC 82.4 82.3 83.3 79.9 81.9 79.7 81.4 79.4
MiddlePTW 52.6 51.2 53 53.1 51.4 49.4 49.4 56.5
MoteStrain 93.4 95.6 93.1 92 95 88.1 74.8 88.2
OliveOil 91.1 90 93.3 86.7 92.6 93.3 93.3 93.3
OSULeaf 98.9 90.1 90.9 90.6 95.9 88.6 88.4 89.5
PhalangesOC 81.3 79 79.7 81.5 82.5 78.7 76.6 80.7
Plane 100 100 100 100 100 99 99 100