Machine learning applications for apprenticeships and employment services
Machine learning could make all the difference in markets like apprenticeships, employment services or education. But how is ML applied to practical scenarios?
Knowing which apprentices are are a higher risk of dropping out - or which job seekers will gain or stick out employment - has been difficult for service providers to understand in the past.
That’s why apprentices completion rates, for example, are still languishing at just above 50%.
However, we are getting smarter with data. With AI and machine learning, we are exploring the potential of predictive intelligence to use data to understand and learn better than ever before.
Here's how it could add value in employment services and apprenticeships.
Machine learning and Australian employment services
Machine learning could have a powerful impact in the Australian employment services market.
Employment services providers have a critical job – they support job seekers into meaningful employment. The way the system is structured, payments are delivered to providers throughout the job seeker lifecycle, with job placement and job retention being key outcomes.
Could ML be used to augment the role played today by case managers and job coaches?
A first reaction might be that, no, case managers are on the frontline with their job seeker participants. They are sitting there across from them, looking into the whites of their eyes, and making intuitive assessments based on experience that only humans are capable of.
But getting that right can also be challenging. Employment consultants have varying degrees of experience and expertise, and there is some time to achieve competence and mastery. Yet at the same time, large-scale market changes (like Workforce Australia) happen at short notice.
To design an efficient public service that scales, could ML support better human services?
ReadyTech has built ML capabilities that can be used to maximise key outcomes for each job seeker based on all available data, so resources are appropriately allocated. This model can predict when an intervention will improve the likelihood of placement and employment.
The model can make use, make sense of and learn from different types of data. This includes structured data like system entries against key job seeker outcomes and payment milestones, as well as unstructured data like case manager notes, using powerful neural networks.
When productised, employment services providers will gain the use of a powerful feature that facilitates automated caseload segmentation, optimisation and workload allocation, to support the efficient allocation of resources inside an employment service provider.
For job seekers it means meaningful employment outcomes and ultimately better lives.
Machine learning and the future of apprenticeships
The apprenticeships market has long faced problems with completion rates. ReadyTech has pioneered the development of a proof-of-concept ML project in the apprenticeships space, where deep learning models were created to predict apprentice risk of dropping out.
Our models used native fields in the data set (for example the post code of an apprentice's employer) with synthetic fields we were able to create (like distance from home to work). We also made use of qualitative fields, where analysing unstructured data in mentor notes was used to find meaning in tone and sentiment from a recurrent neural network.
The results? Our project established the key predictive factors that can predict drop out risk.
These include the size of the employer, the location the apprenticeship is undertaken, the distance from the workplace or work site, and the number of peers and supervisors they are interacting with in the workplace, just to name a few of the predictive factors available.
This was used to create an individual risk score with a predictive accuracy of greater than 85%.
Through this project, important lessons have been learned for providers of apprenticeship services, including that targeted mentoring is more effective at preventing drop outs, and using drop out risk and engagement enables us to better categorise and target individuals.