Human Activity Recognition (HAR) with mobile and wearable devices has been deeply studied in the last decades. Research groups working on this topic evaluated their proposed methods mostly on public datasets. However, most of the existing datasets only include inertial sensor data, while it is well-known that additional context data (e.g., semantic location) has the potential to significantly improve the recognition rate.

Only a few datasets for context-aware HAR are publicly available, and their annotations were mostly self-reported in the wild by the subjects involved in data acquisition. This method harms the quality of annotations, thus discouraging the application of supervised models.

DOMINO is a public dataset for context-aware HAR that includes 25 users (wearing a smartphone and a smartwatch) performing 14 activities. During data acquisition, the mobile devices recorded both inertial and high-level context data while our team monitored the quality of the self-reported annotations.

Please cite this article if you use DOMINO in your research: 

Arrotta, L., Civitarese, G., Presotto, R, & Bettini, C. (2023). DOMINO: A Dataset for Context-Aware Human Activity Recognition using Mobile Devices. In 2023 24th IEEE International Conference on Mobile Data Management (MDM). IEEE.


Download the DOMINO Dataset


Inertial and Contextual Data

Accurate Annotations

DOMINO contains about nine hours of data collected by monitoring 25 different subjects.

DOMINO includes a combination of inertial sensor data from smartphones and smartwatches and contextual information about the surroundings of each subject. 

Among contextual data, we include environment type (indoors/outdoors), semantic place (e.g., at home, at the workplace), speed, weather conditions, nearby public transportation routes, height variation, and the environment's audio level.

DOMINO includes the ground truth labels about 14 activities: brushing teeth, cycling, elevator down, elevator up, lying, moving by car, running, sitting, sitting on transport, stairs down, stairs up, standing, standing on transport, and walking.





The recognition of Activities of Daily Living (ADLs) in smart-home environments is crucial for complex health-care systems that continuously monitor the behavior of fragile elderly subjects in their homes. For instance, the sequence of ADLs performed by a subject and their execution modalities may reveal early symptoms of cognitive decline.

A limitation of most of the existing ADLs datasets is that they only include data from single-inhabitant settings, where only one subject is living in the home. This scenario is actually realistic considering the large amount of elderly subjects living alone in their homes. However, multiple subjects may live in the same home (e.g., married couples of elderly subjects, an elderly and her caregiver, a whole family). At the same time, the main publicly available multi-inhabitant datasets considered only environmental sensors for data collection.

MARBLE is a multi-inhabitant ADLs dataset that combines both smart-watch and environmental sensors data. MARBLE includes sixteen hours of ADLs considering scripted but realistic scenarios where up to four subjects live in the same home environment 



Please cite this article if you use MARBLE in your research: 

Arrotta L., Bettini C., Civitarese G. (2022) The MARBLE Dataset: Multi-inhabitant Activities of Daily Living Combining Wearable and Environmental Sensors Data. In: Hara T., Yamaguchi H. (eds) Mobile and Ubiquitous Systems: Computing, Networking and Services. MobiQuitous 2021. Lecture Notes of the Institute for Computer Sciences, Social Informatics and Telecommunications Engineering, vol 419. Springer, Cham.


Download the Marble Dataset



Multiple Sensing Sources

Accurate Annotations

MARBLE contains data collected in a simulated smart home, where up to 4 residents perform activities both jointly and independently in different scenarios.

MARBLE includes sixteen hours of ADLs. Twelve volunteers participated in data collection.

MARBLE includes a combination of environmental and wearable sensors data. 

Among environmental sensors, we include magnetic sensors to detect open/close of drawers and doors, mat (pressure) sensors to detect when residents are sitting on chairs/sofa, plug sensors to detect the usage of home appliances.

MARBLE includes the ground truth labels about 13 ADLs and the rooms in which they are performed. The ADLs are answering the phone, clearing the table, cooking, eating, entering home, leaving home, making a phone call, preparing a cold meal, setting up the table, taking medicines, using pc, washing dishes, and watching tv

MARBLE also includes the ground truth about the association between each environmental sensors event and the subject that triggered it. In this way, researchers can evaluate the performance of novel data association strategies.