Machine Learning in Java - Second Edition

Correlation coefficient

Correlation coefficient (CC) compares the average of prediction relative to the mean, multiplied by training values relative to the mean. If the number is negative, it means weak correlation; a positive number means strong correlation; and zero means no correlation. The correlation between true values X and predictions Y is defined as follows:

The CC measure is completely insensitive to the mean and scale and less sensitive to the outliers. It is able to capture the relative ordering, which makes it useful for ranking tasks, such as document relevance and gene expression.