Table 9 Performance of models using standardized view in the cross-dataset scenario, using leave-one-dataset-out strategy.

From: A benchmark for domain adaptation and generalization in smartphone-based human activity recognition

Model

Time

Frequency

KH

MS

RW-T

RW-W

UCI

WDM

Mean

KH

MS

RW-T

RW-W

UCI

WDM

Mean

KNN

58.3%

49.8%

38.1%

36.6%

34.6%

43.4%

43.5%

61.1%

81.4%

66.4%

58.8%

59.6%

72.2%

66.6%

Random Forest

54.3%

60.3%

47.6%

42.6%

65.8%

59.0%

54.9%

62.8%

82.8%

67.0%

69.6%

79.5%

71.4%

72.2%

SVM

52.1%

62.4%

49.5%

49.9%

66.5%

55.7%

56.0%

52.8%

80.6%

72.4%

68.3%

67.7%

69.6%

68.6%

CNN (1D)12

63.3%

79.0%

71.7%

68.0%

81.1%

70.3%

72.2%

66.2%

84.6%

71.4%

69.7%

77.2%

73.9%

73.9%

CNN (2D)12

61.4%

70.2%

61.9%

67.8%

70.7%

60.3%

65.4%

66.1%

83.6%

73.9%

69.9%

75.9%

73.3%

73.8%

CNN PF34

61.9%

67.2%

65.0%

66.1%

73.0%

54.9%

64.7%

71.9%

82.8%

70.5%

70.6%

78.3%

74.1%

74.7%

CNN PFF34

63.2%

66.3%

64.8%

67.5%

74.4%

56.0%

65.4%

70.3%

84.3%

69.5%

70.4%

77.6%

74.1%

74.4%

ConvNet13

63.9%

65.6%

47.3%

61.2%

70.1%

53.9%

60.3%

69.7%

85.6%

75.0%

71.3%

81.8%

79.1%

77.1%

IMU CNN14

54.2%

62.5%

42.9%

48.7%

64.8%

59.4%

55.4%

70.4%

85.6%

68.5%

71.3%

78.8%

74.7%

74.9%

IMU Transf.14

63.1%

58.5%

35.7%

57.5%

62.6%

59.8%

56.2%

67.5%

84.0%

67.7%

68.1%

76.4%

73.8%

72.9%

MLP (2 Layers)

55.7%

71.8%

54.9%

55.1%

68.7%

60.9%

61.2%

74.2%

83.9%

70.5%

65.0%

74.6%

73.2%

73.5%

MLP (3 layers)

53.3%

73.6%

54.2%

56.3%

67.3%

59.7%

60.8%

77.9%

85.4%

73.5%

67.2%

75.5%

75.2%

75.8%

ResNet15

58.5%

68.0%

41.1%

66.9%

76.8%

57.4%

61.4%

62.9%

79.8%

66.8%

65.5%

74.1%

69.1%

69.7%

ResNetSE67

60.4%

68.8%

47.1%

68.1%

73.3%

54.2%

62.0%

58.2%

77.5%

68.4%

66.9%

74.7%

67.1%

68.8%

ResNetSE-567

49.0%

67.0%

49.6%

66.2%

72.7%

51.7%

59.4%

65.4%

81.1%

67.5%

66.3%

75.0%

70.6%

71.0%

Max

63.9%

79.0%

71.7%

68.1%

81.1%

70.3%

72.2%

77.9%

85.6%

75.0%

71.3%

81.8%

79.1%

77.1%

  1. The best results for each dataset and for each domain (time and frequency) are highlighted in bold. Mean column represents the average performance of the model in the datasets.