Table 4.

Comparison of the sex-classification results in the literature (semiautomatic) and those of the best-performing automatic approach presented in this paper.

Work	Age	Meas.	Method	N	Test	Acc ${}^{(a)}$	Acc ofthis work${}^{(a)}$ ${}^{(b)}$
Saini et al. [31]	23–65	DB (5)	DFA	116	–	0.802	0.881 (+ 7.9%)
Giles [32]	21–75	DB (9)	DFA	265	TT	0.850	0.871 (+ 2.1%)
Steyn and Işcan [10]	–	DB (5)	DFA	81	–	0.815	-${}^{(c)}$
Dayal et al. [36]	25–69	DB (6)	DFA	60	CV	0.839	0.847 (+ 0.8%)
Pokhrel and Bhatnagar [9]	–	DB (4)	DFA	79	–	0.829	-${}^{(c)}$
Abualhija et al. [21]	21–45	OPG (3)	LoR	50	TT	0.800	0.857 (+ 5.7%)
Franklin et al. [11]	18–70	3DS (38)	PDM+LoR	225	CV	0.831	0.878 (+ 4.7%)
Lin et al. [8]	21–70	3D CT (10)	DFA	240	LOO	0.879	0.871 (− 0.8%)
This work	18–70	OPG (96)	RC	935	TT	0.878

(a) Shape parameters and centroid size were used, as they yielded the best results (Table 5)

(b) The accuracy was calculated for the same age range than original publications. The percentage differences were also reported

(c) The accuracy could not be calculated for the same age range, as the original work did not report this information

Meas.: Measurements. Meas. legend: DB: dry bone; 3DS: 3D scanner; CT: computed tomography. Method legend: DFA: discriminant function analysis; LoR: logistic regression; PDM: point distribution model; RC: ridge classification. N: sample size. Test approach legend: TT: train-test; CV: cross-validation; LOO: leave-one-out. Acc: accuracy