Identifying and Measuring Playful Parenting Using Machine Learning
About this project
|January 2023 - December 2025|
|Principal Investigator||Prof Mark Tomlinson|
|Co-Investigators||Marguerite Marlow, Dr Caspar Addyman & Dr Daniel Statmate (Goldsmiths, UK)|
This innovative project aims to create a new objective measure of playful parenting by applying existing machine learning techniques to videos of caregiver-infant interactions. This project will be led by Stellenbosch Stellenbosch University.
Assessing responsive caregiving is key to evaluating the effectiveness of early interventions. But progress in this regard at any level of scale (such as at population level) has been slow as it is expensive and time consuming. The measures that do exist are often proprietary, require expensive training, skilled professionals to establish reliability, and need to be completed in centre- based settings.
The playful parenting construct – 1) actively engaging; 2) meaningful; 3) iterative; 4) socially interactive; and 5) playful – all fall firmly within the ambit of responsive caregiving (and creating early learning opportunities). Creating an automated and objective measure of interaction and playful parenting has the potential to be a game changer. This is a complex field, and while attempts have been made previously to automatically code facial expressions and interpersonal synchrony, progress has been slow. Recently, some progress has been made using a deep learning model to generate wire- frames from videos that capture information such as gaze to extract novel synchrony measures.
This project will continue to build on this technology, using data collected from the GPI platform to train and test the software.
The project will be divided into three streams:
- Develop a gold standard coding scheme for playful parenting
- Develop software that automates the scoring of caregiver-child interactions
- Tool validation
The GPI platform will provide a vitally important field-testing site. A broad range of data will already be collected, including 5-minute parent-child interaction videos. A selection of these videos could be manually scored by trained coders to establish ground truth and parallel coded with the new algorithm-based screening tool.
Measuring responsive caregiving in parenting and early childhood development (ECD) interventions has always been hampered by cost, time and personnel needed to do the coding. An algorithm-based screening tool has the potential to radically change this. It would not eliminate cost –interactions still need to be filmed – but costs would be markedly lower as well as having scores that are more objective. While not all caregivers and children in a population-based survey could be filmed in a 5-minute interaction video, the ubiquity of mobile phone use, coupled with an algorithm-based screening tool has the potential to revolutionise responsive caregiving measures at a population level.