Aggregation Functions in Researching Connections Between Bio-Markers and DNA Micro-arrays

In this contribution a combination of diverse types of regression models in the case of microarray datasets is considered. There is a proposition of ensemble algorithm where datasets for training constituent regression models are created in a deterministic way. Moreover, aggregation functions are used here to combine the output values of the constituent regression models. Proposed model were tested with Regression Tree, K-Nearest Neighbours and Support Vector Regression as individual models. Some known families of aggregation functions defined on arbitrary [a, b] interval are applied and compared. The applied aggregations are arithmetic mean, exponential mean, olimpic aggregation, arithmetic-min average, arithmetic-max average and median. Moreover, the proposed approach is compared with the bagging regression model with the optimized parameters based on Grid Search. Typical measure such as RMSE is applied to evaluate the proposed ensemble model. The proposed ensemble regression model with some set of parameters outperforms significantly the corresponding single models which was proved using a statistical test.