Machine learning under uncertainty in medical informatics

Aims and scope

The prevalence of uncertain data is based on several factors such as imprecise measurement systems, natural variations over time and linguistic expression. Medical data in particular is associated with different sources of uncertainty, yet the importance of uncertain data in the healthcare area is not sufficiently highlighted. A straightforward example is when patients in a survey are asked about the number of smoking times in a day for which they cannot give an exact answer, they may be able to provide a range and how it changed over a rough time frame. A similar case is encountered with patients’ drinking habits: research shows that general practitioners (GPs) do not believe the patients’ drinking levels, as it often happens that the patients lie to their GPs or they cannot keep track of their drinking. The uncertainty can be indirect as well, such in the case of the random distribution of blood sugar levels in patients that had a meal prior to the test being conducted. Special consideration are examples with missing values which could be either missing at random (MAR) or missing not at random (MNAR). When medical practitioners are not sure about the boundaries of different levels of risk it adds an additional component in the list of uncertainty factors. The last but not the least, uncertainty is always associated with human thinking and judgment which plays an important role in medical decision-making.

The way of handling uncertainty is not well-defined in the machine learning models up to date. Most of the machine learning techniques have been developed based on the assumption of having complete and certain data. The methods that handle missing data by using the mean value per feature are not applicable with categorical data and can add a lot of noise in the model or skew the data/class labels. Therefore, the need for more exploration on uncertain data and developing appropriate machine learning technique for handling uncertainty rises. This requires the development of machine learning techniques which are robust against the changes in the dynamic environments. Consequently, the research community’s interest on applying machine learning techniques in the medical area and developing robust models in terms of uncertainty has increasingly grown in recent years.

This special session aims to showcase the current research (from both academia and industry) on data mining and machine learning applications in medical and healthcare areas and expand these researches’ boundaries by considering uncertainty. It will attract researcher and practitioners who encounter real-case problems while analysing medical data. Special attention will be devoted to handle missing values treatment, imbalance data, pattern recognition and capturing uncertainty in medical applications.

The extended version of the selected papers of this special issue would be considered for publication in journal Frontiers in Digital Health

Topics

The main topics of this special session include, but are not limited to, the following:

• Robust data mining in medical informatics
• Reliable prediction for imbalance medical data
• Medical decision-making under uncertainty
• Capturing uncertainty in medical informatics
• Multiple imputation techniques for medical missing values
• Semi-supervised learning in health informatics
• Information loss mitigation in data mining
• Machine learning using fuzzy logic
• Optimisation (especially robust optimisation) models in machine learning
• Machine learning using uncertainty quantification

Special session organisers

Dr Hadi A. Khorshidi, The University of Melbourne, Australia (hadi.khorshidi@unimelb.edu.au)

Prof Uwe Aickelin, The University of Melbourne, Australia (uwe.aickelin@unimelb.edu.au)

Dr Goce Ristanoski, The University of Melbourne, Australia (gri@unimelb.edu.au)

Important dates

15 January 2020

Paper submission due

15 March 2020

Notification of acceptance

15 April 2020

Camera-ready deadline

15 April 2020

Author registration deadline

Submission process

Submit the paper to this special session:

Submit your paper

Then, you need to determine the special session in “Main research topic” from the drop-down list under “Cross-disciplinary Special Sessions” as SC12.

The accepted papers of this special session will be published in the proceeding of “2020 International Joint Conference on Neural Networks”.

Biographies

Dr Hadi A. Khorshidi

Hadi A. Khorshidi is a Research Fellow in the School of Computing and Information Systems at the University of Melbourne. He has extensive research experiences in medical data-mining, optimisation, machine learning, and uncertainty. He completed his PhD in Applied and Computational Mathematics at Monash University where his thesis was “System Reliability Optimisation via Uncertainty Quantification”. Before joining Melbourne, he worked as a Senior Data Analyst in the Institute of Safety, Compensation and Recovery Research (ISCRR) where he conducted several health-related data-mining projects for Victorian governmental organisations Transport Accident Commission and WorkSafe. He has published more than 35 peer-reviewed journal articles and conference papers. He is an investigator in a joint research grant-awarded project between the Universities of Melbourne and the University of Manchester called Technology for Access to Law. He is a member of editorial boards in International Journal of Quality and Reliability Management and The TQM Journal. He has been a member of the technical committee in IEEE Industrial Engineering and Engineering Management (IEEM) and Australasian Data Mining (AusDM) Conferences. He has served as a reviewer in high-quality journals and conferences such as IEEE Transactions on Fuzzy Systems and IEEE International Conference on Fuzzy Systems (FUZZ-IEEE).

Prof Uwe Aickelin

Uwe Aickelin is an internationally renowned academic in the fields of Artificial Intelligence, Optimisation and Data Mining and since 2018 the Head of the School of Computing and Information Systems at the University of Melbourne.  His specific expertise is in the modelling stages of decision problems with a focus on robust methods to overcome uncertainty. Typical application areas of his work are Decision Support and Optimisation in Health Informatics and Security. He has authored over 200 papers in leading international journals and conferences (Google citations 10000+, H-index 54). Since 2007 he has been an associate editor of the leading international journal in his field (IEEE Transactions on Evolutionary Computation). He also served for many years as a strategic adviser for Artificial Intelligence to the UK Research Councils and Government. He has conducted several projects for the UK Government Communications Headquarter (GCHQ) and the UK border force. In total, he has won over 10 million AUD of research funding as Chief Investigator and over 40 million AUD as Co-Investigator.

Dr Goce Ristanoski

Goce Ristanoski is a Research Fellow in the School of Computing and Information Systems at the University of Melbourne. He has extensive research experiences in applied machine learning and data mining, working both in research and industry background. He completed his PhD in machine learning at The University of Melbourne, specialising in time series. Before joining The University of Melbourne, he worked at Data61 (CSIRO), Australia’s leading company if the area of AI research. His work involved projects that applied machine learning models in telecommunications, cybersecurity, retail industry, medical data. He has also served as a committee member of several international conferences.