Predicting the Development of Type 1 Diabetes With Machine Learning Based Expert System
There are numerous factors affecting the development of Type 1 Diabetes (T1D), and we do not know how these factors interact. According to current research the development of T1D is usually a long process and there are many changes in metabolism during that process. The data we are gathering in the DIPP project gives us a unique opportunity to examine both the environmental factors and changes in the metabolism since birth.
Every person participating in the study produces several data variables at every visit. Also, the samples taken on these visits are analysed in laboratory and produce further data variables, which tell us about the metabolism and autoantibodies. Due to the enormous amount of data it is usually difficult for the human mind to notice the combination of effects from several factors.
We can use bioinformatics to process and analyse these vast quantities of data. Our goal is to develop an automated machine learning based expert system that both manages the data of study subjects and simultaneously develops with predictive analytics such mathematical models that describe the development of T1D and would give early warning in rise of the progression risk. Vast amount of data gives us the opportunity to find early signals of T1D development, and we can compare these signals to the data of study subjects and predict whether the risk of progression is rising.
The expert system that we are developing learns and develops prediction models based on constantly accumulated data. Since the predictions produced by this system are probabilities, and thus ambiguous, it is mandatory that human experts evaluate these predictions and give feedback to the system. Then the system can use this feedback to learn and develop these prediction models even further.