Exploration of Black Boxes of Supervised Machine Learning Models: A Demonstration on Development of Predictive Heart Risk Score

Table 2

Extraction of relative feature weights using ANN and linear SVM.

Features	Artificial neural network		Linear support vector machine
	Sum of input feature contribution	Relative feature weights (%)	Original feature weights	Relative feature weights (%)
Age groups	0.836	11.944	1.085	8.828
Parental history of CVDs	0.401	5.735	0.330	2.683
Self-reported general stress	0.360	5.138	0.732	5.957
Consumption of high salty foods	0.429	6.128	1.024	8.338
Low fruit consumption	0.666	9.517	0.832	6.773
Physical inactivity	0.562	8.024	1.167	9.499
High fried foods/trans fats	0.404	5.770	0.857	6.976
Abdominal obesity	0.406	5.794	1.046	8.512
Diabetes mellitus	0.446	6.371	0.978	7.961
Hypertension	0.737	10.527	1.386	11.283
Smoking history	0.635	9.067	0.969	7.887
Low vegetables consumption	0.588	8.394	0.894	7.273
Red meat/poultry consumption	0.531	7.591	0.987	8.030
Total	7.000	100.000	12.286	100.000