Sandhya Indurkar

Math foundations

The Normal Distribution: Bell Curves, Mean, and Standard Deviations

Normal distribution bell curve with sigma bands

The idea

Many operational metrics cluster around a center with symmetric tails: delivery times, order values, measurement error. The normal distribution models that shape with two parameters: mean mu and standard deviation sigma.

The empirical rule gives fast mental math: about 68% of mass within one sigma, 95% within two, 99.7% within three. Z-scores count how many standard deviations a value sits from the mean, linking this curve to feature scaling and outlier checks.

Normal is a working model, not a law of nature. Check fit before you trust tail probabilities.

Example: bell curve, mean, and sigma bands

The normal density clusters around mu. Sigma sets spread. Shaded bands show the 68-95-99.7 empirical rule in action.

Most orders cluster near the mean. Sigma sets how wide the bell spreads.

+/-1 sigma

68%

+/-2 sigma

95%

+/-3 sigma

99.7%

mu=45.068%95%

Mean 45.0 min, sigma 8.0 min. About 68% fall within +/-1 sigma, 95% within +/-2, and 99.7% within +/-3 when the normal model fits.

The math

Probability density function

f(x) = (1 / (sigma sqrt(2 pi))) exp(-(x - mu)^2 / (2 sigma^2))

Bell-shaped density centered at mu. Sigma stretches or tightens the curve. Total area under the curve equals 1.

Empirical rule

68% within mu +/- sigma, 95% within +/- 2 sigma, 99.7% within +/- 3 sigma

Quick coverage bands for roughly normal data. Useful for SLA bands and anomaly thresholds when you have not plotted the full distribution yet.

Z-score link

z = (x - mu) / sigma

Standardizes any normal variable to mean 0, sigma 1. Connects this post to feature scaling and to asking how extreme one observation is.

A simple application

When you set an alert at mu + 3 sigma, you are assuming normality and targeting roughly 0.15% tail mass on the high side. Plot the histogram first. Heavy tails or skew mean the normal approximation will miss real outliers or over-flag routine noise.

Use the normal model for quick capacity planning and error budgets. Pair with variance-spread and percentile posts when the data is not symmetric.