InclusiveHAR: A Smartphone-Based Dataset for Human Activity Recognition Across Diverse Physical Abilities

InclusiveHAR is a multivariate time-series dataset designed for human activity recognition using wearable smartphone sensors, with a focus on individuals with physical disabilities. The dataset includes data collected from 20 participants, consisting of 10 non-disabled individuals and 10 individuals with disabilities.

Data were recorded using an iPhone 14 Pro smartphone worn vertically in a waist pouch. Motion and location signals were captured from the device’s internal sensors, including accelerometer, gyroscope, magnetometer, motion sensors, and GPS-derived data, at a sampling rate of 50 Hz. The SensorLog app was used to lock the rate at 50 Hz.

Participants performed six daily activities: walking, standing, sitting, jogging, ramp ascent, and ramp descent. For safety reasons, the ramp ascent and ramp descent activities were conducted on an inclined ramp with an 8% slope. Each activity was repeated multiple times to ensure consistent recordings. Additionally, "walking" labels for wheelchair users refer to manual propulsion.

The dataset is organized as a time-series table in which each row corresponds to a single time-step. It contains 30 feature columns representing sensor signals, along with an activity label column, a binary disabled indicator where 0 denotes non-disabled participants, and 1 denotes participants with disabilities, and a UserID column identifying each participant.

InclusiveHAR is intended to support research on machine learning and deep learning methods for human activity recognition, particularly in healthcare, rehabilitation, and assistive technology applications. All participants provided informed consent, and data collection was conducted under professional supervision with strict attention to safety and privacy. Additionally, detailed descriptions of individuals with disabilities have been provided as supplementary material.