Measuring Bias in AI Models: An Statistical Approach Introducing N-Sigma