The Problem

Inclusive Design is becoming more important with the ageing of the world’s population and improvements in medical care. Designers are required to respond to this population shift by executing an Inclusive Design process to produce practical inclusive consumer products across a range of sectors. There remains a need for a better understanding of how data on human capability variation across populations can support the inclusive approach. Current user data is fragmented and there is a lack of integrated data with proven predictive value.
A collaborative research project spearheaded by Dr. Umesh Persad (Associate Professor, UTT Design and Manufacturing Engineering, Engineering Design and Product Development Group) and researchers at the Cambridge Engineering Design Centre sought to better understand the capability variation in ageing and disabled populations by the novel application of new unsupervised machine learning techniques.

What was done

Topological Data Analysis (TDA) (unsupervised machine learning) was used to extract insight from a unique capability dataset. The aim was to explore the global shape and sub-groupings (clusters of profiles) of people using data collected from the Cambridge Better Design Pilot Study of 362 people from across England and Wales. The unique dataset contained capability variables describing the age, gender, vision, hearing, cognition and motor function of participants.

The TDA method combines the mathematical field of Topology with Machine Learning and Visualisation. It is built on the principle that data has a multidimensional shape, and this shape conveys meaning. It is a geometric method to detect patterns and shapes within the data. By recognising these shapes and patterns in the data, important features and groupings could be identified. The advantage of TDA is that it can detect patterns missed by traditional multidimensional methods such as PCA, MDS and cluster analysis.
The data was imported and analysed in AYASDI Platform, a TDA software tool for analysis and visualisation (www.ayasdi.com). The TDA method requires no prior assumptions allowing the data to speak for itself. A network visualisation of 14 clusters was produced and clustered using a community auto grouping algorithm. The network algorithm, based on Louvain Modularity optimisation, operates on the topological model’s graph structure. It tries to find the best grouping of nodes that have high intragroup connectivity and low intergroup connectivity, resulting in highly connected clusters.

Results
The resulting topological network demonstrated the global shape of the sample and distribution of sensory, cognitive and motor capability across the sample. The TDA network was automatically grouped into 14 distinct clusters, and distinguishing features of each cluster was extracted. In each age group, there was a spread of capability from single minor capability loss to multiple capability loss. In addition, the attitudes and experience of people in each age group can provide designers with the data that they need to create designs that are usable, accessible and easy to learn.
The variation exhibited by the data underscores the importance of Inclusive Design approaches when designing for the wider population. Supporting and capturing the richness of user diversity in design approaches, methods and tools will become more important as populations age and healthcare improvements enable longer life. The results demonstrate the value of applying TDA to analyse and visualise user capability data, and it is proposed that the cluster descriptions could be used for developing empirically based design tools such as personas for Inclusive Design.

Impact of Results
The results demonstrate the usefulness of the TDA approach using machine learning and network visualisation to explore and extract insight from user capability data. The machine learning approach shows promise for application in future Ergonomics/Human Factors studies that capture large multivariate data sets. The Better Design pilot study points the way to future large-scale data collection efforts with multiple sensory, cognitive and motor variables. Given that analysis and visualisation tools such as TDA will make it easier to see the global structures inherent in data, it will encourage a move to methodologies that allow the data to “speak for itself” and build new theoretical and practical insights. Further work on the Better Design data will focus on relating user capabilities to rated product difficulties with an eye to developing predictive models for analytical product evaluation.

Publications
This work was presented at the Cambridge Workshop on Universal Access and Assistive Technology 2018 and published as a Book Chapter:
Persad U., Goodman-Deane J., Langdon P.M., Clarkson P.J. (2018) Exploring User Capability Data with Topological Data Analysis. In: Langdon P., Lazar J., Heylighen A., Dong H. (eds) Breaking Down Barriers, Springer, 41 – 50.

Acknowledgements
Dr. Umesh Persad would like to acknowledge the contribution of AYASDI Inc. (www.ayasdi.com) to this research project by providing access to its web-based machine learning software, AYASDI Platform, via an Academic Collaboration Agreement.