ePrints.FRI - University of Ljubljana, Faculty of Computer and Information Science

Prediction of a kidney disease using machine learning methods

Romana Koprivec (2012) Prediction of a kidney disease using machine learning methods. EngD thesis.

Download (2266Kb)


    A group of researchers announced a competition for making diagnostic models that predict intensity of obstructive nephropathy. In this thesis we have developed a regression model which can predict the level of the illness according to the molecular profile. This model is intended to predict two target values: pelvic diameter and differential renal function. We wanted to improve the prediction models with the knowledge about potential connections between biological levels. We have implemented a procedure with the help of the published data about connections between attributes and Mann-Whitney test, which connects two different databases on two different groups of samples. The lowest relative root mean squared error (RRMSE) has been achieved using locally weighted regression. RRMSE of the method has been lower on connected databases than on the original databases. The result on test dataset has improved as well. The best estimated regression methods on train and test dataset have differentiated because of extraordinary small number of samples. Some models had higher RRMSE on train dataset; however, they achieved better results on test dataset. Nevertheless, with the best model according to RRMSE on train dataset we have exceeded the best-published result on the competition for 81%.

    Item Type: Thesis (EngD thesis)
    Keywords: machine learning, regression, bioinformatics, obstructive nephropathy
    Number of Pages: 34
    Language of Content: Slovenian
    Mentor / Comentors:
    Name and SurnameIDFunction
    doc. dr. Zoran Bosnić3826Mentor
    Link to COBISS: http://www.cobiss.si/scripts/cobiss?command=search&base=50070&select=(ID=00009072980)
    Institution: University of Ljubljana
    Department: Faculty of Computer and Information Science
    Item ID: 1645
    Date Deposited: 28 Mar 2012 19:54
    Last Modified: 12 Apr 2012 17:21
    URI: http://eprints.fri.uni-lj.si/id/eprint/1645

    Actions (login required)

    View Item