Comparing Machine Learning Methods to Improve Fall Risk Detection in Elderly with Osteoporosis from Balance Data

Table 4

Specificity (S), sensitivity (Se), positive predictive value (P), and negative predictive value (N) measures obtained with the different classifiers using FSMC’s variables (FSMC), Weka’s variables (Weka), and all variables (all). AdaBoost (meta), Naïve Bayes (Bayes), IBk (KNN), LibSVM (SVM), and Random Forest (tree) classifiers were built using balanced datasets with the oversampling method. Measures in bold = best classifiers’ performance.

			Open eyes					Closed eyes					Merge
			Meta	Bayes	KNN	SVM	Tree	Meta	Bayes	KNN	SVM	Tree	Meta	Bayes	KNN	SVM	Tree
Oversampling data	All	P	0.70	0.78	0.86	0.50	0.82	0.73	0.35	0.91	0.48	0.85	0.68	0.71	0.97	0.34	0.87
		N	0.53	0.38	0.65	0.96	0.72	0.42	0.72	0.65	0.96	0.73	0.48	0.56	0.66	0.99	0.77
		S	0.39	0.42	0.24	0.27	0.23	0.42	0.46	0.22	0.28	0.21	0.42	0.37	0.19	0.34	0.18
		Se	0.61	0.58	0.76	0.73	0.77	0.58	0.54	0.78	0.72	0.79	0.58	0.63	0.81	0.66	0.82
	FSMC	P	0.62	0.32	0.79	0.60	0.74	0.67	0.29	0.84	0.46	0.83	0.71	0.31	0.76	0.57	0.83
		N	0.52	0.78	0.63	0.60	0.68	0.56	0.75	0.67	0.88	0.70	0.51	0.76	0.63	0.82	0.66
		S	0.43	0.45	0.29	0.40	0.29	0.38	0.48	0.25	0.33	0.24	0.39	0.47	0.31	0.31	0.26
		Se	0.57	0.55	0.71	0.60	0.71	0.62	0.52	0.75	0.67	0.76	0.61	0.53	0.69	0.69	0.74
	Weka	P	0.69	0.69	0.86	0.69	0.80	0.66	0.25	0.89	0.79	0.80	0.75	0.37	0.87	0.51	0.81
		N	0.48	0.44	0.58	0.81	0.71	0.46	0.80	0.62	0.65	0.69	0.50	0.73	0.63	0.92	0.73
		S	0.42	0.44	0.28	0.25	0.25	0.44	0.48	0.24	0.28	0.25	0.38	0.45	0.25	0.29	0.23
		Se	0.59	0.56	0.72	0.75	0.76	0.56	0.52	0.76	0.72	0.75	0.63	0.55	0.75	0.71	0.77