Decision support systems for aircraft engine product development (2013-2020)
Analysis and improvement of Random forests (2012-)
Identification of connected burglaries (2012-2015)
Previous Research Projects
Automatic discovery of surgery indicators (2009-2012)
This is joint work with Dr. Marie Persson. Sweden, and most other developed countries, struggle with increasing healthcare costs. Hospital treatment and surgery are among the most expensive forms of treatment. Usually, a patient goes to see a general practitioner. If needed, the general practitioner refers the patient to specialist care at the hospital. The estimation of patient demand to surgery is done at the hospital. We hypothesize that an estimation could be carried out earlier if predictions about whether to perform surgery or not could be made earlier (perhaps as early as during the referral stage). We use data mining technologies on data from the hospital to generate models from which we may extract surgery indicators. Such indicators could then be used as a basis for defining new, structured referrals to be predict surgery at an earlier stage. This work is funded by Blekinge County Council and Blekinge Institute of Technology.
Predicting the risk of hospitalization for the elderly (2009-2010)
Elderly people over the age of 80 is the fastest growing group in Sweden. Due to this growth, the proportion of people with several chronical diseases, multiple pharmaceuticals and disabilities increases. Hospitalization accounts for a large amount of the cost of healthcare. This project investigates if it is possible to predict the risk of future hospitalization, given information about a patient's diagnoses, pharmaceuticals. We are analyzing a database consisting of such information for 406,272 patients from 2006 and a database featuring the number of days of hospitalization during 2007 for the same patients. Supervised learning algorithms will be used to generate risk predictors. This project is funded by Blekinge County Council.
Metric-based learning (2006-2010)
Supervised learning algorithms are usually designed to optimize a certain goal metric, e.g. accuracy or information gain. However, the goals of real-world applications vary substantially and it may be possible to tailor some existing algorithms to efficiently solve specific problems by replacing their learning/goal metric with a metric that represent the explicit goals of the problems.
Diagnosis and Cause Analysis of Acute Coronary Syndrome (2008-2010)
We performed a pilot study, in which we investigated the possibility to classify the severity of an Acute Coronary Syndrome (ACS) by generalizing from a subset of patients featured in a data set that includes both environmental and biological factors. The complete data set features 843 patients, (601 men and 242 women) aged 30-74 years, with ACS. These patients were consecutively recruited in the Carlscrona Heart Attack Prognosis Study (CHAPS), a study population that entailed consecutive patients admitted to the coronary intensive care unit at Blekinge hospital Karlskrona between 1992-1996. This project was funded by Blekinge Research Council.
Spyware Prevention by Mining End User License Agreements (2007-2010)
This work was initiated by me, Martin Boldt, Prof. Paul Davidsson, and Dr. Andreas Jacobsson (currently at Malmö University) in 2007. The use of the Internet has become mainstream and downloadable software applications are available everywhere. The last decade has witnessed an enormous growth of spyware, i.e., software that report information about users to e.g. marketing companies without the users' informed consent. Spyware vendors usually want to avoid being prosecuted and so they often mention in the End User License Agreement (EULA), which is shown during the installation of an application, that the software contains spyware. However, this information is given in a way that makes it hard to understand or even spot. In this project we investigate the use of Data Mining Algorithms to detect spyware by analyzing EULAs. This project is funded by Blekinge Institute of Technology.
Evaluation of Classifiers and Supervised Learning Algorithms (2004-2008)
This work was a continuation of my Master's thesis, which was presented in June 2003, and was carried out for my PhD thesis. I investigated current evaluation approaches, methods, and metrics for supervised concept learners and classifiers. The project was funded by Blekinge Institute of Technology.