Feature overview table
Note that all catch22 features are statistical properties of the z-scored time seriesāthey aim to focus on properties of the time-ordering of the data and are insensitive to the raw values in the time series.
In the table below we give the original feature name [from the Lubba et al. (2019) paper], and a shorter name more suitable for use in feature descriptions.
Features are also (loosely) categorized into broad conceptual groupings.
# | Feature name | Short name | Category | Description |
---|---|---|---|---|
1 |
|
| 5-bin histogram mode | |
2 |
|
| 10-bin histogram mode | |
3 |
|
| Positive outlier timing | |
4 |
|
| Negative outlier timing | |
5 |
|
| First crossing of the ACF | |
6 |
|
| First minimum of the ACF | |
7 |
|
| Power in lowest 20% frequencies | |
8 |
|
| Centroid frequency | |
9 |
|
| Error of 3-point rolling mean forecast | |
10 |
|
| Change in autocorrelation timescale after incremental differencing | |
11 |
|
| Proportion of high incremental changes in the series | |
12 |
|
| Longest stretch of above-mean values | |
13 |
|
| Longest stretch of decreasing values | |
14 |
|
| Entropy of successive pairs in symbolized series | |
15 |
|
| Histogram-based automutual information (lag 2, 5 bins) | |
16 |
|
| Time reversibility | |
17 |
|
| First minimum of the AMI function | |
18 |
|
| Transition matrix column variance | |
19 |
|
| Wang's periodicity metric | |
20 |
|
| Goodness of exponential fit to embedding distance distribution | |
21 |
|
| Rescaled range fluctuation analysis (low-scale scaling) | |
22 |
|
| Detrended fluctuation analysis (low-scale scaling) |
And in some cases, in which scale and spread of the raw time-series values may be relevant to class differences, the two simple distributional moment features (using the catch24
flag in the software implemenations) can be added:
Feature name | Short name | Description |
---|---|---|
|
| Mean |
|
| Standard deviation |
Last updated