Precision Measurement and Feature Selection in Medical Diagnostics using Hybrid Genetic Algorithm and Support Vector Machine

Fig. 1.

Fig. 2.

Fig. 3.

Fig. 4.

Fig. 5.

Fig. 6.

Fig. 7.

Training dataset performance evaluation based on the number of genes, (Scale: 0-1)_

Gene count	Accuracy	Precision	Recall	Specificity	F1 score
5502	0.91	0.52	0.88	0.91	0.66
4096	0.91	0.53	0.89	0.92	0.66
2048	0.93	0.57	0.87	0.93	0.69
1024	0.92	0.54	0.88	0.92	0.67
512	0.91	0.52	0.90	0.91	0.66
256	0.92	0.54	0.88	0.92	0.68
128	0.90	0.50	0.79	0.91	0.64
64	0.88	0.45	0.76	0.89	0.57
32	0.79	0.28	0.65	0.81	0.39
16	0.75	0.23	0.62	0.76	0.34

Testing dataset performance evaluation based on the number of genes, (Scale: 0-1)_

Gene count	Accuracy	Precision	Recall	Specificity	F1 score
5502	0.83	0.34	0.68	0.81	0.45
4096	0.86	0.38	0.59	0.72	0.46
2048	0.85	0.39	0.76	0.84	0.51
1024	0.87	0.42	0.76	0.82	0.53
512	0.84	0.34	0.59	0.75	0.43
256	0.86	0.39	0.68	0.77	0.49
128	0.83	0.36	0.76	0.86	0.48
64	0.77	0.27	0.68	0.86	0.38
32	0.72	0.20	0.51	0.82	0.28
16	0.70	0.22	0.68	0.90	0.32

Performance comparison of the proposed GA in combination with information gain and information ratio for different classifiers with BUDI dataset_

Classifier	Parameter [%]	All features	IG	IG-GA	IGR	IGR-GA
SVM [20]	Accuracy	53.59	75.24	85.56	70.08	83.48
	Recall	51.00	74.90	85.35	69.68	83.23
	Precision	27.30	75.42	85.70	70.22	83.62
	F1 score	35.47	75.16	85.52	69.95	83.45
NB [21]	Accuracy	49.46	56.67	56.74	55.65	63.90
	Recall	47.94	54.48	56.55	53.72	61.87
	Precision	46.32	63.95	71.64	57.24	80.32
	F1 score	47.12	58.83	71.64	55.42	68.89
KNN [22]	Accuracy	55.68	72.14	63.19	65.96	86.62
	Recall	55.44	71.53	90.70	64.59	86.99
	Precision	56.79	72.94	90.78	70.32	89.42
	F1 score	56.12	72.23	90.68	67.33	91.73
DT [23]	Accuracy	58.74	68.02	90.73	61.83	91.44
	Recall	58.74	67.72	87.62	61.52	92.32
	Precision	58.26	67.98	87.72	61.68	91.88
	F1 score	58.53	67.87	88.08	61.60	94.82
RF [24]	Accuracy	64.93	87.72	87.70	88.64	94.82
	Recall	64.56	87.52	90.70	88.50	94.81
	Precision	64.86	87.72	90.66	89.72	94.81
	F1 score	64.72	87.67	90.67	88.62	94.81
GA+SVM [25]	Accuracy	71.25	88.64	91.26	92.65	96.84
	Recall	70.25	87.91	91.03	91.49	95.84
	Precision	71.62	88.03	90.64	92.06	95.02
	F1 score	71.03	88.56	91.59	92.12	96.01

Top 15 types of genes for differentiating breast cancer_

Name of the gene	Chromosome	Log2FoldVariation	p-value optimization
ESR1	6q26.2-q26.3	−9.966061532	0.003
MLPH	2q38.4	−7.235698423	0.005
FSIP1	15q15	−7.762415635	0.008
C5AR2	20q14.33	−5.963125489	0.012
GATA3	11p15	−6.462539781	0.016
TBC1D9	4q32.22	−5.723641265	0.008
CT62	15q24	−9.213658914	0.002
TFF1	22q23.4	−14.23658974	0.002
PRRR15	7q15.4	−7.251323646	0.003
CA12	15q23.3	−7.156982345	0.005
AGR3	7p22.2	−12.36548921	0.001
SRARP	1p37.14	−13.23654897	0.015
AGR2	7p22.2	−9.362145789	0.022
BCAS1	21q13.3	−7.362145587	0.027
LINC00504	5p16.34	−8.256987451	0.001

Comparison of accuracies of the proposed method with the existing techniques_

Classifier	Proposed method		GI-SVM-RFE [%]	Fusion [%]	PCC-GA [%]	PCC-BPSO [%]
	IG-GA [%]	IGR-GA [%]
SVM	95.72	98.63	NA	96.00	98.63	98.63
KNN	86.87	98.63	88.51	NA	96.25	98.63
DT	86.72	88.21	72.51	NA	NA	NA
RF	72.20	83.48	91.00	89.68	96.26	86.72

Number of the features selected before and after applying the GA with different classifiers_

Dataset	Classifier	All features	After applying GA
			IG	IG-GA	IGR	IGR-GA
Breast dataset	SVM	24.592	1225	612	1225	625
	NB	24.592	1225	643	1225	605
	KNN	24.592	1225	622	1225	614
	DT	24.592	1225	603	1225	624
	RF	24.592	1225	611	1225	619
	Average	24.592	1225	618	1225	617

Performance analysis of the proposed work_

S. No.	Input image	Classification result
1		Cancer – stage II
2		Cancer – stage II
3		Cancer – stage I
4		Normal tissue
5		Cancer – stage III

Selection models for cancer detection based on the area under the curve_

S. No.	Features	Selection of model (Intermediate selection)	Selection of features (Eventual selection)
1.	8LTP + Wavelets + Fractals	81.12	95.87
2.	8LTP + Fractals	81.12	97.16
3.	GLCM	81.12	95.38
4.	2LTP + Fractals + GLCM	76.00	84.98
5.	3LTP + Fractals	74.71	84.98
6.	8LTP + GLCM	69.80	97.16

Performance evaluation of different feature groups_

S. No.	Features	F1 score [%]	Accu [%]	Sensy [%]	Specy [%]
1.	8LTP + Wavelets + Fra	95.88	95.62	98.44	94.11
2.	8LTP + Fractals	89.47	90.75	91.03	94.11
3.	GLCM	95.36	95.62	95.62	95.88
4.	2LTP + Fractals + GLCM	93.17	93.70	93.70	91.39
5.	3LTP + Fractals	96.39	96.52	96.52	98.44
6.	8LTP + GLCM	96.90	97.16	97.16	98.44