A state-machine model used to predict the future behavior of a stochastic process using the absolute minimum amount of historical memory. It groups past events into causal states that have the exact same probability of producing future outcomes. Engineers use these models to find hidden structures in complex, noisy time-series data where standard regression or standard control charts fail.
Standard autocorrelation isn't catching the pattern in this noise, so let's try building an epsilon-machine to see if there's a hidden state structure.
Building an epsilon-machine is overkill for basic process control, but it excels at diagnosing highly complex, non-linear system failures.
Related terms
Autocorrelation
Correlation between successive values in a time series, which breaks the independence that standard control charts assume.
Regression
Fitting an equation that predicts a response from one or more input variables, usually by least squares.
SPC
Statistical process control: using control charts and run rules to tell routine process noise from real changes.