public class AutomaticPeriodIdentification
extends Object
Finds number of period of time series. Should work for arbitrary time series.
Algorithm:
1. If a time series is trend stationary trend is removed using first differences
2. Computes auto correlation function for the length of the series
3. Finds highest correlation of the series and set period its index
4. Checks if significant(@see +-CORRELATION_MIN/MAX_CHANGE) correlations exists at following periods
e.g. found period was 4, so it checks if at index 8, 16, 20 exists significant correlations
5. If failed go to step 3.
- Author:
- Pavol Loffay