Multi-criteria neural network estimation of correlation coefficients for processing small samples of biometric data

Abstract

Background. Over 120 years, Pearson’s criterion has been widely used to calculate correlation coefficients. Unfortunately, its use generates significant errors in calculating the correlation coefficients on small samples. The purpose of this research is to reduce these errors that occur with small samples by increasing the complexity of data processing. Materials and methods. We consider the reduction of errors in calculating the correlation coefficients by using the large artificial neural networks, trained to predict the values of the correlation coefficients from the relative position of small sample points. Results. The combined use of the classical Pearson’s formula and the neural network calculation of correlation coefficients can significantly increase the level of confidence in the neural network calculations. Conclusions. It is noted that training samples of a neural network computer can be obtained from software random data generators and can be large. This allows us to hope for a significant increase in the accuracy of calculating the correlation coefficients.

Keywords

• neural network computing
• pearson’s criterion
• correlation coefficient
• neural network analysis
• data independence criteria
• small samples