Award-winning probabilistic framework sets new accuracy standard

The Data Science and Advanced Analytics conference recognised TAIAO researcher and AI and Data Science Associate Professor Yun Sing Koh, Ben Halstead and her team of researchers for their study on existing probabilistic framework methods and the creation of a new method.

As the world generates an unprecedented amount of data, the need for accurate and efficient machine learning systems have become extremely important – including probabilistic frameworks (PFs) which predict future possibilities through data statistics.

However, a challenge that machine learning systems like PFs face is concept drift – where observed patterns in data are no longer the same as they were in the past.

When concept drift occurs, models that were previously accurate may become outdated and/or inaccurate, and new models need to be retrained. This makes it hard for machine learning and artificial intelligence systems to make predictions and decisions.

Through researching PFs, the team found that current PFs, although highly accurate with complex data streams, can still produce inaccurate information because of how they read – or don’t read, past data.

“Simply put, existing PFs only review past data inputs when the current information is deemed ‘irrelevant’, rather than at each data input.” says Dr Koh.

“So, in the event of a concept drift, existing PFs cannot guarantee each prediction is based on the most relevant information until the data falls below a certain relevancy mark.”

From this finding, the team created a new PF method, ‘SELeCT’, that continuously reads past and present data to ensure the PF prediction is accurate.

“Because SELeCT finds the most relevant data (whether from past or present data) at each data input, it can notice concept drifts and create predictions based on the most relevant data as it’s happening rather than after the fact.”

The team evaluated SELeCT against existing methods and found that their predictions were consistently more accurate than other PFs.

“It’s a great feeling for myself and the team to have put so much work into this PF, see such great results and then be awarded because of it,” says Dr Koh.

To learn more about SELeCT and its creation process, click here.