Table 2. Comparison between the models using the testing dataset.

ML models were able to predict the state of the fore and hind legs. RF surpassed all other learning algorithms in all respects and scenarios. In some cases, RF did not have significant superiority over KNN. Accordingly, KNN was the second most efficient algorithm among all the characteristics and scenarios.

Model	Accuracy	Kappa	P-value κ	Sensitivity	Specificity
RF	0.8846	0.7693	<2.2e−16	0.8232	0.9463
KNN	0.8754	0.7509	3.238e−16	0.8013	0.9499
C50Tree	0.6469	0.294	0.001603	0.5746	0.7195
Boost	0.6035	0.207	0.09968	0.5995	0.6075
NNET	0.5667	0.1335	0.01852	0.5619	0.5716
LDA	0.563	0.1258	2.343e−05	0.5986	0.5272
GLM	0.5624	0.1246	5.043e−05	0.5971	0.5275
SVM	0.5603	0.1202	3.248e−05	0.653	0.4671
NB	0.5411	0.0816	<2.2e−16	0.6984	0.3832