sit-stand_XXX_v1.0.0

sit-stand-baseline_v1.0.0

First version of the sit-stand baseline test.

This test records labeled sit-to-stand and stand-to-sit transitions under ideal conditions with minimal noise and a normal transition pace (~1.5 seconds per transition). It establishes a clean baseline for evaluating sit-stand detection algorithms.

Test Changes

• First version of the test (no changes)

Performance

Test	Algorithm 1
sit-stand-baseline_v1.0.0	DELTA_Y_THRESHOLD Precision Recall TP FP FN 0.05 0.842 1.000 16 3 0 0.10 0.842 1.000 16 3 0 0.15 0.842 1.000 16 3 0 0.20 0.889 1.000 16 2 0 0.30 0.941 1.000 16 1 0 0.40 0.941 1.000 16 1 0 0.50 0.941 1.000 16 1 0 0.60 0.941 1.000 16 1 0 0.70 0.941 1.000 16 1 0

The default threshold of DELTA_Y_THRESHOLD = 0.20 sits in the flat-recall region (2 files, 16 ground-truth events).

Attached Files

• baseline_v1.0.0.csv — IMU CSV recording — result of running sit-stand-baseline_v1.0.0
• baseline_v1.0.0.json — Annotation file (state labels) — result of running sit-stand-baseline_v1.0.0

These files were recorded with the intention of evaluating sit-stand detection. Other activities present in the files (e.g. initialization, stable sit/stand periods) are annotated and may be used to evaluate false positive rates.

---

sit-stand-slow_v1.0.0

First version of the sit-stand slow test.

This test records labeled sit-to-stand and stand-to-sit transitions performed deliberately slowly (~5 seconds per transition). It is designed to capture edge cases where transitions are not abrupt and the resulting gyroscope signal is weaker.

Test Changes

• First version of the test (no changes)

Performance

Test	Algorithm 1
sit-stand-slow_v1.0.0	DELTA_Y_THRESHOLD Precision Recall TP FP FN 0.05 0.842 1.000 16 3 0 0.10 0.842 1.000 16 3 0 0.15 0.842 1.000 16 3 0 0.20 0.842 1.000 16 3 0 0.30 0.842 1.000 16 3 0 0.40 0.889 1.000 16 2 0 0.50 0.882 0.938 15 2 1 0.60 0.875 0.875 14 2 2 0.70 0.867 0.812 13 2 3

The default threshold of DELTA_Y_THRESHOLD = 0.20 achieves perfect recall but begins to trade off above 0.40 as slow transitions produce smaller acc_y changes (2 files, 16 ground-truth events).

Attached Files

• slow_v1.0.0.csv — IMU CSV recording — result of running sit-stand-slow_v1.0.0
• slow_v1.0.0.json — Annotation file (state labels) — result of running sit-stand-slow_v1.0.0

These files were recorded with the intention of evaluating sit-stand detection under slow transition conditions. Other activities present in the files (e.g. initialization, stable sit/stand periods) are annotated and may be used to evaluate false positive rates.

---

sit-stand-noisy_v1.0.0

First version of the sit-stand noisy test.

This test records labeled sit-to-stand and stand-to-sit transitions with intentional movement noise injected during seated and standing periods (e.g. leaning, repositioning, weight shifting, partial stand attempts). It is designed to evaluate algorithm robustness against false positives from realistic in-session movement.

Test Changes

• First version of the test (no changes)

Performance

Test	Algorithm 1
sit-stand-noisy_v1.0.0	DELTA_Y_THRESHOLD Precision Recall TP FP FN 0.05 0.421 1.000 16 22 0 0.10 0.457 1.000 16 19 0 0.15 0.500 1.000 16 16 0 0.20 0.516 1.000 16 15 0 0.30 0.516 1.000 16 15 0 0.40 0.571 1.000 16 12 0 0.50 0.571 1.000 16 12 0 0.60 0.640 1.000 16 9 0 0.70 0.696 1.000 16 7 0

Recall is perfect across all tested thresholds; precision is lower than baseline due to noisy movement generating spurious gyroscope peaks that pass the classification gate (2 files, 16 ground-truth events).

Attached Files

• noisy_v1.0.0.csv — IMU CSV recording — result of running sit-stand-noisy_v1.0.0
• noisy_v1.0.0.json — Annotation file (state labels) — result of running sit-stand-noisy_v1.0.0

These files were recorded with the intention of evaluating sit-stand detection under noisy conditions. Noise activity occurs within labeled sit and stand intervals and is not separately annotated as distinct states.

---

sit-stand-walking_v1.0.0

First version of the sit-stand walking test.

This test records labeled sit-to-stand and stand-to-sit transitions interspersed with walking segments. It is designed to introduce realistic false positives from walking activity and evaluate the algorithm's ability to suppress them using walking detection.

Test Changes

• First version of the test (no changes)

Performance

Test	Algorithm 1
sit-stand-walking_v1.0.0	DELTA_Y_THRESHOLD Precision Recall TP FP FN 0.05 0.591 0.929 13 9 1 0.10 0.650 0.929 13 7 1 0.15 0.722 0.929 13 5 1 0.20 0.722 0.929 13 5 1 0.30 0.765 0.929 13 4 1 0.40 0.812 0.929 13 3 1 0.50 0.812 0.929 13 3 1 0.60 0.812 0.929 13 3 1 0.70 0.812 0.929 13 3 1

One ground-truth event is consistently missed across all thresholds; the remaining false positives are transitions detected adjacent to walking segments that the soft-exclusion margin does not suppress (2 files, 14 ground-truth events).

Attached Files

• walking_v1.0.0.csv — IMU CSV recording — result of running sit-stand-walking_v1.0.0
• walking_v1.0.0.json — Annotation file (state labels) — result of running sit-stand-walking_v1.0.0

These files were recorded with the intention of evaluating sit-stand detection in the presence of walking. Walking segments are labeled as walk in the annotation file and can be used to assess walking suppression independently from overall detection accuracy.

---

Results Graphs

These results were ran on the sit_stand.py module in CFL_Wearable_ADRD_Biometric_Lib.

sit-stand-sit_v0.0.0

sit-stand-sit_v0.0.0

The first version of the sit-stand-sit test.

This test includes tap detection, a period of sitting, a period of standing, another period of sitting, and then tap detection.
The purpose of standing and sitting in this test is to increase the variance in the heart rate data throughout the length of the test.

Test Changes

• First Version of the test (no changes)

Performance

Test	Algorithm 1
sit-stand-sit_v0.0.0	Score Plot Scores Value r 0.3225 rho 0.7197 tau 0.5497

Attached Files

• cyhr8.txt - Cyton File - result of running sit-stand-sit_v0.0.0
• ebhr8.csv - Unparsed EmotiBit file - result of running sit-stand-sit_v0.0.0

These files were recorded with the intention of investigating heart rate. As such, other biometric signals found in the files may not be existent or reliable.