Confidence Interval

What is a Confidence Interval?

A confidence interval is the range of values needed to match a confidence level for estimating the features of a complete population. Confidence intervals are usually reported in the context of a margin of error, though they are two unique values. While separate, confidence intervals are closely connected to confidence levels.

The minimum confidence level is set by the machine learning human trainer, usually at 95%, though sometimes lower or higher depending on whether an unsupervised or supervised learning technique is being used.

The interval is how large a range of values you need to reach that confidence level that the sample’s results would reflect the entire population’s features. 

The primary factors influencing how “tight” the confidence interval can be are the size of the sample, the confidence level, and the variability within the sample. For example, a larger and more random sample will produce a higher confidence level estimate for the total population.

Confidence Intervals Versus Confidence Levels

Confidence levels are the likelihood of the results being true for the total population and expressed in percentage form, for example 95%. So if you reproduced that sample or experiment repeatedly, 95% of the time you should see the same results if you counted the total population. 

Confidence intervals are the results of the samples or experiments. For example, if you take a sample with a 99% confidence level, you’ll likely have a large high/low estimate range of values. If you lowered confidence to 90%, the range would be much smaller.