Cancer Classification using Gene Expression

DNA mikročipovi moderna su tehnologija za mjerenje razine ekspresije gena koja primjenu pronalazi kod razotkrivanja tumora na temelju izmjerenih razina ekspresija gena. U ovom radu proučavana je primjena strojnog učenja u analizi podataka prikupljenim pomoću tehnologije DNA mikročipova s ciljem postavljanja dijagnoze raka prostate na temelju dobivenih podataka. Stroj potpornih vektora i logistička regresija pokazali su se najboljim rješenjem klasifikatora za problem DNA mikročipova, dok je metoda rekurzivne eliminacije obilježja s varijabilnim korakom koja koristi stroj potpornih vektora za dodjelu težine obilježjima dala bolje rezultata nego metode koje koriste algoritam nasumične šume za računanje težina obilježja.DNA microarrays are a modern technology used for measurement of gene expression levels. The method has been widely used for making a distinction between a tumor cell and a normal cell based upon the measured gene expression levels. In this thesis, a prostate tumor dataset was used for testing commonly used machine learning algorithms for the DNA microarray problem. Support vector machine and logistic regression have proven to be the best choice for this type of data, especially in combination with the recursive feature elimination with variable step size algorithm which assigns the weights of features using a support vector machine estimator. The recursive feature elimination with variable step size has outperformed methods for reducing the number of features which use a random forest algorithm as an estimator