Table 2. A comparison between experimental and predicted irreps to independently validate the classification models.

RP oxides	Experimental irrep	Predicted irrep	Prediction accuracy (in %)
CaSrRuO4 (ref. 74)	P4	P4	60
LaSrFeO4 (ref. 75)	φ	φ	100
LaSrCoO4 (ref. 76)	φ	φ	100
NdSrCoO4 (ref. 76)	φ	φ	100
GdSrCoO4 (ref. 76)	φ	φ	100
LaSrCrO4 (ref. 77)	φ	φ	100
YCaCrO4 (ref. 77	(η0,η1)	(η0,η1)	80
YSrCrO4 (ref. 77)	(η1,η2)	φ	0
SmCaCrO4 (ref. 78)	(η0,η1)	(η0,η1)	100
LaCaFeO4 (ref. 79)	(η0,η1)	(η0,η1)	80
Ca2CrO4 (ref. 80)	P4	P4 and (η0,η1)	40
NaDyTiO4 (ref. 16)	(η1,η1)	(η1,η1)	100
NaSmTiO4 (ref. 16)	(η1,η1)	(η1,η1)	100
NaHoTiO4 (ref. 16)	(η1,η1)	(η1,η1)	100

Prediction accuracy (in %) is the ratio of the number of trees that correctly predicted the irrep label to the total number of trees (=5) used for prediction. All experimentally reported compounds have disordered A-site arrangement. In Ca₂CrO₄, our classifier predicts with 40% confidence that both P₄ and (η₀,η₁) labels are equally likely and experimentally, P₄ is observed.