Alzheimer's disease is an increasingly pressing problem in the modern world, as the world's population is aging. This insidious neurodegenerative disease causes dementia and affects the daily life of both -- the diseased and their caregivers. Diagnosing this disease is currently possible only after the death of the affected person. Researchers are striving to develop a method for predicting this disease using different types of data.
In this master's thesis we are trying to improve prediction of Alzheimer's disease using data fusion. To this end, we propose an early attribute integration method. Our method takes common attributes of two sets and builds models to extend the attributes from original set with the missing attributes of the other set. We test the our implementation of the proposed method on multiple sets containing data of various modalities. We compare classification models built on the fused datasets and on the original datasets. Our implementation of the method does not significantly improve the models. However, in some cases, improvement is noticeable and can nevertheless indicate the potential of data fusion for increasing the classification accuracy.
|
