Inside Dutch Social Housing: Smart Sensors Connecting Daily Life With Energy Use

Making social housing more energy efficient matters because many residents have little money to spare when energy bills rise. The paper cites research reporting that 75% of energy-poor households in the Netherlands live in social rented housing, which makes up nearly one-third of the country’s housing.

A heating bill records how much energy a household uses, leaving the daily life behind those numbers largely out of sight: rooms warming up before breakfast, appliances running through the afternoon, doors opening as people move around their homes. Connecting those details can help researchers understand what comfortable, affordable housing needs to look like.

The study 'A multisensor dataset of energy use and activity in Dutch social housing,' appearing in npj Scientific Data, introduces HOMESENSE, a dataset bringing together energy measurements, indoor conditions and activity signals from 11 social-rental homes in Limburg, the Netherlands. Yu Sheng and colleagues at Open Universiteit followed the homes across 366 consecutive calendar days, from 26 November 2024 through 26 November 2025, creating a resource for exploring how everyday routines, heating and weather fit together.

1. Why these homes deserve a closer look

Making social housing more energy efficient matters because many residents have little money to spare when energy bills rise. The paper cites research reporting that 75% of energy-poor households in the Netherlands live in social rented housing, which makes up nearly one-third of the country's housing.

Homes in these neighbourhoods often have similar designs and are managed by a few housing associations, making it easier to organise improvements across several properties. Studying everyday conditions inside these homes can help researchers explore ways to lower energy use and keep residents comfortable.

HOMESENSE brings together a full year of information about appliances, room conditions, heating settings, movement and available electricity and gas readings. Few existing datasets offer this combination for social housing. The 11 homes had similar two-floor layouts, with living areas usually downstairs and bedrooms upstairs. This makes comparisons between the homes useful, though the group is too small to represent all Dutch households

2. What the sensors recorded behind the front door

Living-room sensors tracked temperature, humidity, carbon dioxide, air pressure, light and a device-based air-quality rating. Smart radiator valves recorded heating settings, motion and door sensors tracked activity, and smart plugs measured electricity used by appliances such as televisions, refrigerators and washing machines.

Eight homes provided electricity readings, and seven provided gas readings. The electricity data showed power taken from or sent to the grid, rather than total household use. Outdoor temperatures from Maastricht helped researchers compare energy use with the weather.

All readings were organised into 10-minute time slots, although sensors reported at different intervals. Quality labels show whether each value was recorded, estimated, reconstructed after a meter reset or missing. Some gas readings cover longer than 10 minutes. Motion sensors show detected movement, so a lack of movement does not necessarily mean nobody is home.

3. The patterns that emerged across the year

Across 2,463 complete household-days from seven homes, the average within-home correlation was −0.869, indicating a strong association between colder days and higher gas consumption. Gas use rose sharply in the morning, with the weekend increase happening somewhat later, and another rise appeared in the late afternoon. Weekday electricity imports showed local peaks around 8 a.m. and between 5 p.m. and 7 p.m.; the weekend profile was flatter.

The motion records showed activity building through the morning and remaining more frequent during daytime and evening hours than overnight. These profiles describe signals from monitored locations, without identifying residents or confirming their schedules.

Living-room carbon dioxide concentrations tended to be higher during hours with detected movement. Across 91,291 household-hours, the average within-home correlation was 0.109, a modest relationship that supports consistency between the two sensor streams without making either one a reliable headcount.

A separate test home provided an additional check on the measurement system. Digital readings closely matched utility billing records over nine complete months, with aggregate differences below 0.14% for gas and imported and exported electricity. That comparison checked the recording process in the test property; it did not independently validate every participating household.

4. What researchers can build from this resource

HOMESENSE could help researchers study heating routines, indoor conditions and differences in energy use between homes. It could support work on moving electricity use to times when demand on the power system is lower. Readings from individual appliances could help researchers estimate which devices contribute to a home's electricity use using data organised into 10-minute time slots.

The dataset contains about 41 million measurement entries. Around 69.3% come from recorded observations, 22.3% are estimates between available readings, and 8.4% are missing. A tiny share was reconstructed to handle meter resets. Each entry has a quality label so researchers can understand where it came from and decide whether to use it.

The researchers left incomplete days or hours out of the energy summaries because counting missing readings as zero would make energy use appear lower than it was. Sensors received no further calibration after installation.

The team did not collect information about residents' ages, jobs or children, or the buildings' ages and energy ratings. The data cannot explain differences linked to these details, and the study does not prove that a particular heating routine lowers bills. It is most useful for exploring patterns within the participating homes.

All participants agreed to the monitoring, and the project received ethical approval. The public files use household codes instead of personal details and identify locations only as being within Limburg. Sharing readings in 10-minute time slots reduces some privacy risks, though it cannot remove them completely.

Researchers can access 11 household data files, processing code and guides for loading the data and repeating the published analyses. The original, unprocessed household records are restricted, so users cannot rebuild the entire dataset from the raw source.