Application of Machine Learning Approaches for Classifying Sitting Posture Based on Force and Acceleration Sensors
Table 1
Classification accuracy of the different pattern recognition algorithms.
Classification algorithm
Classification accuracy
Support Vector Machines (SVMs)
Linear kernel function
70.1%
Quadratic kernel function
78.5%
3rd polynomial kernel function
77.7%
4th polynomial kernel function
79.8%
5th polynomial kernel function
75.9%
6th polynomial kernel function
72.2%
RBF kernel ()
82.7%
Multinomial Regression (MNR)
87.8%
Boosting
Subspace
47.6%
72.4%
AdaBoostM2
85.2%
78.8%
LPBoost
82.1%
62.3%
TotalBoost
75.1%
68.5%
RUSBoost
82.0%
27.4%
Bag
81.8%
90.4%
Neural Networks (NN)
Feedforward NN
Feedeforwardnet
trainlm(22)
90.4%
trainscg(23)
88.3%
traingd(22)
39.1%
traingdm(22)
37.2%
traingdx(23)
77.5%
Patternnet
trainlm(17)
89.2%
trainscg(15)
85.1%
traingd(24)
25.3%
traingdm(25)
27.6%
traingdx(18)
71.7%
Fitnet
trainlm(21)
89.8%
trainscg(18)
88.4%
traingd(25)
40.0%
traingdm(24)
36.3%
traingdx(24)
78.4%
Cascade-forwardnet
trainlm(22)
90.0%
trainscg(22)
78.7%
traingd(23)
45.2%
traingdm(21)
39.3%
traingdx(22)
41.7%
Radial Basis NN
newrbe
72.5%
newgrnn
69.5%
newpnn
69.5%
Random Forest (RF)
NVarToSample = 2
90.5%
NVarToSample = 3
90.7%
NVarToSample = 4
90.9%
NVarToSample = 5
90.2%
Combination: Boosting, NN, RF
90.8%
Only the results with the highest classification accuracy for the different hidden layer sizes (NN) as well as for the different scaling factors (SMV with RBF kernel, Radial Basis NN) are listed. The highest accuracies are marked in bold for every category. Weak learner for the corresponding Boosting method. Neural network training function with the hidden layer size in brackets that showed the highest classification accuracy. The highest classification accuracy shown throughout variation of the scaling factor of the radial basis function .
