Invited by the Department of Immigration and Border Protection, Prof Longbing shared his views on deep behaviour analytics to a group of over 30 senior analytics professionals from various government departments as a means of motivating and inspiring professional practice in government analytics. The talk introduces the theory of “behaviour informatics”, a new area proposed by Longbing, and illustrates latest techniques and real-world showcases developed by Longbing’s team for deeply understanding individual and group behaviours, non-occurring behaviours, high impact behaviours, high utility behaviours, and collective behaviours which share coupling relationships. Deep behaviour analytics is very powerful for an in-depth understanding of customer/consumer’s behaviours, interactions between clients and service providers, impact of positive and negative behaviours, business utility of behaviours and more effective response and treatment strategies and actions in broader business and government domains.
Our young PhD candidate, Shameek Ghosh, has also been invited to share his health informatics research at HISA NSW’s 3rd Young Talent Time event. The talk aims to introduce to health professionals the idea of data mining for intensive care units and the various methods. Topics include Prediction of Acute Hypotensive Episodes using Sequential Contrast Pattern Mining: A Study in Critical Care Data Mining. The development of acute hypotension in a critical care patient leads to significant complications in his or her health status. Acute hypotension, characterised by extremely low blood pressure, if left untreated, may rapidly lead to multiple organ damage. Traditional patient monitoring systems trigger an alarm when they detect a fall in blood pressure below a certain threshold, at which point a medical intervention is staged to prevent organ damage. Thus, early detection of impending drops in blood pressure could be of significant importance in improving ICU patient outcomes. In this context, existing patient monitoring alarms can be considered as suboptimal, given the high rate of false alarms and a reactive approach to identifying low blood pressure.
In the study, Shameek presents a novel application for mining informative sequential contrast patterns towards the early prediction of acute hypotensive episodes. The relevant data was extracted from MIMIC-II, which is a freely available (subject to NIH certification) de-identified intensive care unit research database, aggregating high-resolution diagnostic and therapeutic data from a large, diverse population of adult intensive care unit patients since 2000. The results demonstrate the competitiveness of the approach, in terms of both prediction performance as well as interpretability. The importance of this work is associated with the introduction of a powerful sequential pattern mining-based knowledge discovery process for analysing physiological time series data towards making critical care predictions.
