Traditional regression analysis has been a staple in predicting cause-effect relationships. Counltess industries use these methods to predict rare events, from economic issues to cancer research, causality is often a desired result. Given the vast amount of potential variables it sometimes becomes overwhelming to plug them into the analysis. An alternative to the mind-numbing calculus which arises from tradtional methods is needed.
Dr. David Melamed has developed an algorithm to meet this issue. The algorithm esssentially reduces the pool of cases when looking for a rare event. The algorithm uses configurational methods to identify the conditions, or combinations of conditions, that lead to the non-occurrence of rare events. By excluding cases of non-occurrence, the baseline likelihood of occurrence increases. Preliminary analyses (ground truth and simulations) are quite promising. In one set of simulations, it was possible to reduce the pool of possible cases by 60%.
- Accuracy in eleminating non-variables
Dr. Melamed is an Associate Professor of Sociology and is affiliated with the Translational Data Analytics Institute. His research interests are in Social Networks, Group Processes, Computational Modeling, Stratification, and Theory.
TTO Home Page: https://tco.osu.edu/
Name: Andrew Hampton