A team of researchers from the Johannesburg School of Business (JBS) at UJ is advancing jobs and skills forecasting across industries in South Africa in a significant way.
Prof Arnesh Telukdarie heads the team employing big data and AI to deliver next-generation forecasts to industry sectors.
For 2024, the researchers supplied the Chemical Industry Sectoral Training Authority (CHIETA) with the new type of jobs and skills forecast.
On average, about 850 companies submit skills plans to the CHIETA. About 20,000 companies form the sector, of which 92% are small, with fewer than 50 staff.
Previously, the SETAs used to simply forecast for the subsequent year’s jobs, based on the current year’s data using spreadsheets, with only the year-on-year change visible.
It used to take 3 or 4 months per SETA, says Telukdarie.
The new forecast takes a day, with automated AI-based data cleaning, and is done across industry sectors.
The new forecast also incorporates and visualises 8 or 9 years’ data per occupation for a longer-term view, making it possible to identify job trends.
Telukdarie is Professor of Digital Business at the JBS. Mr Mpho Nkadimeng and PhD candidate Mr Xolani Maphisa are the research collaborators.
A feature developed by Nkadimeng enables examining the influences possibly driving new trends.
On the financial side, the JBS forecast database is linked to the SETA’s levy data, CPI and other financial indicators, says Telukdarie.
“That database integrates which companies pay the SETAs levies, how much they pay, and where they are located,” he adds. The output facilitates accurate interpretations driving a better understanding of the link between an occupation’s forecast, the SETA investment in training for it, and growth areas.
For each job forecasted for a SETA, the team also forecasts a skills matrix as required, says Telukdarie. They use AI and big data scraping of millions of jobs of international job sites.
PhD candidate Mr Xolani Maphisa built a semantic matching Large Language Model (LLM) using BERT. He trained the model on a 20,000-point dataset.
Maphisa says they had a total of 3 million data points in their AI database, which is growing every month. Using the model, they could then map the CHIETA jobs to the jobs and skills found in the global job boards.
The BERT semantic matching makes it possible to match a job role called X in the USA with the same role called Y in South Africa, and Z in Asia. The result is a skills and jobs forecast across SETAs.
The forecast includes globally emerging skills for new technologies and new value chains, as well as local context such as CPI growth per sector, concludes Telukdarie.
This article was originally published in the December 2024 issue of the UJ Research and Innovation Magazine, at https://www.uj.ac.za/research-at-uj/uj-research-and-innovation-magazine/.
