Appendix N — chapter4.analysis.statistical_tests
chapter4.analysis.statistical_tests
Provides functions to compute statistical tests to determine the significance of the obtained results.
Functions
| Name | Description |
|---|---|
| compare_distribution_with_zero | Statistically compares the mean of a distribution with 0 using a T-test (normal data) or a W-test (non-normal data). |
| compare_rmse_distributions | Statistically compares the RMSE of two distributions using a two-sample T-test (normal data) or a MWU test (non-normal data). |
| compare_splitting_approaches | Statistically compares the two splitting approaches on smartphone and smartwatch data. More concretely, |
| compute_icc | Computes the Intraclass Correlation Coefficient (ICC) between the TUG results obtained by the system and |
compare_distribution_with_zero
chapter4.analysis.statistical_tests.compare_distribution_with_zero(distribution)
Statistically compares the mean of a distribution with 0 using a T-test (normal data) or a W-test (non-normal data).
Parameters
| Name | Type | Description | Default |
|---|---|---|---|
distribution |
list[float] | array of numbers constituting the distribution. | required |
Returns
| Type | Description |
|---|---|
| float | P-value of the test. A value less than 0.05 indicates a significant difference between the distribution mean and zero. |
| bool | Indicates if the distribution is normal or not and, therefore, which statistical test was used. |
compare_rmse_distributions
chapter4.analysis.statistical_tests.compare_rmse_distributions(errors_df)
Statistically compares the RMSE of two distributions using a two-sample T-test (normal data) or a MWU test (non-normal data). The compared distributions are the inter-subject RMSE of the TUG duration and each subphase.
Parameters
| Name | Type | Description | Default |
|---|---|---|---|
errors_df |
pandas.DataFrame |
DataFrame containing the error in ms of the system measures and the reference method for all subjects. | required |
Returns
| Type | Description |
|---|---|
pandas.DataFrame |
DataFrame containing the tests results of comparing each measure (i.e., TUG duration and subphases). |
compare_splitting_approaches
chapter4.analysis.statistical_tests.compare_splitting_approaches(reports, metrics)
Statistically compares the two splitting approaches on smartphone and smartwatch data. More concretely, determines if there is a significant difference in the accuracy of the models or the F1-score of the TURNING, SITTING_DOWN and TURN_TO_SIT activities.
Parameters
| Name | Type | Description | Default |
|---|---|---|---|
reports |
dict |
Model reports. | required |
metrics |
list[str] | Performance metric to compare. | required |
Returns
| Type | Description |
|---|---|
pandas.DataFrame |
DataFrame containing the statistical test results. |
compute_icc
chapter4.analysis.statistical_tests.compute_icc(system_results, manual_results, labels, icc_type='ICC2')
Computes the Intraclass Correlation Coefficient (ICC) between the TUG results obtained by the system and the reference methods for each TUG measure.
Parameters
| Name | Type | Description | Default |
|---|---|---|---|
system_results |
list[pandas.DataFrame] |
Contains the DataFrames with the measures generated by both system’s configurations. | required |
manual_results |
pandas.DataFrame |
DataFrame containing the measures generated by the reference method. | required |
labels |
list[str] | Text labels associated with the DataFrame in system_results. |
required |
icc_type |
str | Type of ICC to compute. One of: ICC1, ICC2, ICC3, ICC1k, ICC2k, ICC3k. | 'ICC2' |
Returns
| Type | Description |
|---|---|
| pandas.pandas.DataFrame | DataFrame containing the ICC results for each TUG measure and system (C1 and C2). |