Uploading Dataset to IEEE DataPort to Advance Machine Learning Algorithms
Francesca Meneghello, from the University of Padova in Italy worked with a team to collect valuable Wi-Fi wireless sensing data and uploaded the dataset to IEEE DataPort. This data is contributing to the advancement of Wi-Fi-enabled sensors and algorithms to detect human activity.
Advancing Wireless Sensing Technology
Francesca chose to upload the dataset titled “CSI Dataset for Wireless Human Sensing on 80 MHZ Wi-Fi Channels” using IEEE DataPort because it speeds up research, especially for machine and deep learning-based use cases. Learning-based algorithms require a huge amount of data to learn and train. Publishing the dataset to IEEE DataPort is important for replicability and benchmarking.
“A benefit of IEEE DataPort is its visibility, as it is accessible to the entire IEEE community and, through the open access option, by any interested researcher.”
One of Francesca’s research goals was to provide other researchers with everything they need to reproduce the research to address wireless sensing problems and create algorithms that can be implemented in commercial devices.
The strategy can also be replicated using datasets from other researchers that publish data in IEEE DataPort. This would test the performance of the algorithm on different scenarios and acquisition setups. In fact, having datasets associated with the same task but collected by different research teams replicates the real world, where each user installs the commercial device in different environments and under different conditions. In this way, algorithms can be designed and adjusted to hardware and environment changes.
Connecting with IEEE
The dataset was collected to research contactless human sensing solutions using commercial Wi-Fi devices below 7 GHz frequencies. This line of research is being explored by several researchers around the world and has also attracted interest from the industrial community such as IEEE, which established a working group to integrate sensing facilities into Wi-Fi networks. The datasets have been used to validate signal processing and machine learning-based techniques.
The dataset relates to the article “SHARP: Environment and Person Independent Activity Recognition with Commodity IEEE 802.11 Access Points.” The article has been accepted for open-access publication in the IEEE Transactions on Mobile Computing and is currently available under early access, here.
The article presents SHARP, a novel learning-based algorithm for contactless human activity recognition through commercial Wi-Fi devices. The idea is to leverage the way human movements affect Wi-Fi signal propagation to obtain information on the type of activity performed. Once trained, SHARP can identify the activity in different scenarios based on the training data i.e., with different persons, rooms, device positions, and days of inference.
Francesca assessed the performance of SHARP on the dataset she uploaded to IEEE DataPort, obtaining accuracies higher than 95%. In addition to the dataset, Francesca made the code of the implementation publicly available, so that other researchers can replicate the results and use the same technique as a benchmark.