Table 3.
The net change in the performance metrics across five ablation experiments (AE), each of which contrasted a unique pair from the four variants of the KIDS framework.
| Ablation experiment | KIDS (I) | KIDS (II) | KIDS (III) | KIDS (IV) | Problem domain | Performance metric | Mean net change |
|---|---|---|---|---|---|---|---|
| AE 1 | – | – | 1 | F-score | 0.06 | ||
| Se | 0.06 | ||||||
| PPV | 0.09 | ||||||
| 2 | R | 0.18 | |||||
| AE 2 | – | – | 1 | F-score | 0.09 | ||
| Se | 0.09 | ||||||
| PPV | 0.09 | ||||||
| 2 | R | 0.18 | |||||
| AE 3 | – | – | 1 | F-score | (−) 0.01 | ||
| Se | (−) 0.02 | ||||||
| PPV | 0.00 | ||||||
| 2 | R | 0.15 | |||||
| AE 4 | – | – | 1 | F-score | 0.02 | ||
| Se | 0.05 | ||||||
| PPV | 0.00 | ||||||
| 2 | R | 0.15 | |||||
| AE 5 | – | – | 1 | F-score | 0.08 | ||
| Se | 0.07 | ||||||
| PPV | 0.17 | ||||||
| 2 | R | 0.33 |
The net change in the metrics was averaged over five participants (P1–P5). The variants encompassed: KIDS with UMAP sans the enhancement step (I), KIDS with UMAP coupled with the enhancement step (II), KIDS with ADR excluding the enhancement step (III), and KIDS amalgamating ADR with the enhancement step (IV). Each experiment served to discern the value attributed to the design choices that led to the variants. The performance metrics were categorised according to the problem domains: sleep posture change detection (Problem Domain 1) and temporal segmentation of postural inactivity (Problem Domain 2). The ablation analysis revealed three pivotal observations: (1) the consistently superior performance rendered by ADR-based variants regardless of the incorporation of the enhancement step (AE 1 and AE 2), (2) the pronounced benefit of the enhancement step particularly for ADR-based variants (AE 3 and AE 4), and (3) the amalgamated gains accrued from the simultaneous utilisation of ADR and the enhancement step exceeded the individual advantages offered by either one of them (AE 5).