Table 6.

Comparison with state-of-the-art methods for eight-class classification of BreaKHis dataset.

Reference	Method Used	Accuracy Performance (%)
		40×	100×	200×	400×
Bardou et al [23].	SURF features encoded with bag of words (BoW) and classified with SVM.	49.65	47.00	38.84	29.50
	SURF features encoded with locality constrained linear coding and classified with SVM.	55.80	54.24	40.83	37.20
	Deep CNN features (trained from scratch) with KNN classifier on top.	70.48	68.00	70.08	66.38
	Deep CNN features (trained from scratch) with Linear SVM classifier on top.	72.35	67.68	66.45	64.95
This work	Zernike moments, Haralick, and color histogram features all fused and classified with five standalone classifiers.	87.69	89.32	89.82	85.65
	Block-wise fine-tuned VGG-19 model with softmax classifier on top.	98.13	97.39	96.63	94.05