Table 2.
Classification performance of our Stop & Go algorithm that was used to distinguish dwelling intervals (stops) from transit intervals (trips).a
| Performance | Stop & Go classification including motion score | Stop & Go classification without motion score |
| Correct, n/N (%) | 119,614/122,808 (97.40) | 118,865/122,808 (96.79) |
| Balanced accuracy | 0.965 | 0.966 |
| F1-score | 0.975 | 0.966 |
| Stop counts (system/dairy), n | 667/692 | 708/692 |
| Missed stops, n | 26 | 26 |
| Fragmented stops, n | 19 | 43 |
| Trip counts (system/dairy), n | 667/691 | 708/691 |
| Missed trips, n | 33 | 28 |
| Runtime (seconds) | 33.12 | 49.31 |
aOur algorithm can include accelerometer data to further refine results (ie, “motion score”); however, most conventional stop/trip classifiers do not offer such a feature. For better comparability with other systems, we reported results for both with and without accelerometer data.