| ```html
METODOLOGÍA
1. Metolodogía (empleada para la recogida o generación de los datos)
The data come from a prospective observational study conducted as part of the MOTIVA project, funded by the Spanish Ministry of Science. Its objective is to collect inertial motion data and ground truth data in sufficient quantity to design, train and evaluate a set of models and algorithms for the automatic evaluation of older adults' performance when taking an SPPB test and when completing a physical exercise session of the VIVIFRAIL functional training program.
The participants were volunteers recruited by convenience sampling from a population of older adults aged 70 or older, capable of walking with or without assistance, and living in the community. The participants were contacted via the Day Hospital and outpatient clinics of the Geriatrics Department at Getafe University Hospital and Senior Centers and Day Centers in municipalities within Madrid.
- Particiapts met all the following inclusion criteria:
- Being 70 years of age or older.
- Scoring 4 or higher on the SPPB.
- Being able to walk, with or without assistance.
- Participants did not meet any of the following exclusion criteria:
- Being unable or unwilling to give consent.
- Being unable to understand instructions.
- Having any clinical contraindication, as assessed by the investigator.
- Having a diagnosis of any disease that, in the investigator's judgment, contraindicates physical exercise.
Data were collected with a propietary motion capture system. The system consists of:
- five devices equiped with a 6 DoF IMU (accelerometer and gyroscope), local storage capability via a micro-SD card, and Wi-Fi connectivity.
- a mobile app compatible with Android devices for the configuration of each collection session, the synchronization of the devices, and the collection of ground truth labels.
The data were collected according to the following procedure:
- Data are collected while participants perform the SPPB and VIVIFRAIL exercises naturally, with the only requirement being that they wear the five inertial devices in the motion capture system (one device on each wrist, one device on each ankle, and one device on the lower back).
- The procedure started with the participants standing in a neutral position for five seconds for calibration purposes.
- The participants underwent an SPPB test (1. double stand, semi-tandem and tandem balance tests, 2. 3 meters walking test, 3. 5 times sit-to-stand test).
- The participants performed the exercises of the VIVIFRAIL wheel corresponding to their SPPB score (Wheel D if the score is between 10-12 points, wheel C if the score is between 7-9 points, and wheel B if the score is between 4-6).
- The research team members use the mobile app to label different events that occur during the session (start of an exercise, start of a set, end of a repetition).
- Each device stores a 6D inertial signal with a timestamp for each sample in a csv file in their sd card.
- The app stores the list of labels in a txt file in the internal storage of the Android device.
The data were preprocessed to clean them from errors and spurious labels that may have been generated during the data collection sessions.
- Labels corresponding to repetitions and sets marked as invalid were automatically identified. The detected cases were verified through visual inspection of the graphical representation of the signals along with the overlaid labels. When confirmed as invalid, these labels were disabled by manually modifying the corresponding field in the ground truth file.
- The number of sets per session for each exercise was automatically identified. Anomalies (e.g., a different number of sets than expected) were verified through visual inspection of the graphical representation of the signals with overlaid labels. When necessary, the labels for set start and repetition end were adjusted by modifying the corresponding fields in the ground truth file.
- Duplicate labels (e.g., those resulting from a double press) were automatically identified by detecting labels separated by less than 550 ms. When necessary, only one of the duplicates was kept active by modifying the corresponding field or deleting the extra labels from the ground truth file.
- Start and end labels of sets that were excessively early or late were identified through visual inspection of the graphical representation of the signals with overlaid labels. When necessary, these labels were repositioned by modifying the corresponding field in the ground truth file. | 