Fairfield Dolan professors Christopher Huntley, PhD, and Philip Maymin, PhD, hosted a virtual Business Analytics Showcase on March 25, to highlight the work of two distinguished Department of Analytics faculty members: Jie Tao, PhD, and Xin James He, PhD, both instructors in the Master of Science in Business Analytics (MSBA) program.

Dr. Tao’s research, “Making Sense of the Black-Boxes,” examines what he calls the biggest problem in the field of artificial intelligence (AI): the interpretation of the deep neural network models, commonly called black boxes due to their opacity, that are by nature extremely complex.

A lot of companies won’t adopt AI technology because they don’t trust or understand it, said Dr. Tao, who teaches MSBA courses on Python for data science and machine learning, as well as a new graduate-level elective on deep learning. In order to gain acceptance, he noted that a project must have “a very good workflow design, good communication… and explicit and actionable knowledge to guide certain actions [in areas] the business experts care about.”

“Of course, the Holy Grail is actually explaining the models, but these models have very complicated structures [and] once you make them less complicated, you’re actually compromising the performance,” continued Dr. Tao. The challenge, he said, is to build a method of interpretation that can be applied to any model, and to make it relatable to those without a data science background.

As an example of his approach, Dr. Tao used the game theory-based interpretable machine learning method SHAP to create a text classification model called IDeL (Interpretable Deep Learning). His model outlined a way to integrate black boxes and evaluate them, while removing features that cause errors in data interpretation and result in misclassification.

Dr. He, who teaches the MSBA class on forecasting, has focused his research on “The Challenges of Time Series Forecasting,” which has particular relevancy to the financial, marketing, and insurance sectors. Consider, for example, the difficulty in predicting stock and oil prices, consumer behavior, employment rates, and life expectancy.

Accuracy in forecasting is a major issue, he stressed, citing as an example the sudden stock market drop in December 2018 and fluctuations in 2019, followed by the best first-quarter in the last decade. “The S&P 500 ended 2020 up about 15 percent," he said."Yet hedge fund company Bridgewater Associates lost $12 billion.” Warren Buffet’s Berkshire Hathaway Inc. fared a bit better, gaining two percent. He posited that those companies would pay dearly for a more accurate forecasting model.

Using the financial markets as his example and culling data from Yahoo Finance, Dr. He used a decomposition technique model to look at S&P 500 performance over the last 40 years, using the first 20 of those years to predict performance in the second 20. His conclusion: while decomposition models are the “work horses” in this area, “in practice, you can get more accuracy based on historical data.” 

In class, said Dr. He, “we’ve talked about forecasting with R Studio [but] for long-term investors the S&P 500 would beat Warren Buffett and any hedge fund over time.”