Online Program

Fractal Dimension as a Dimensionality Reduction Tool in Gene Expression Analysis (308423)

Keywords: Fractal Dimension, Dimensionality Reduction, Gene Expression Analysis

A fractal dimension helps to measure the complexity of an object by providing a statistical index of complexity as a ratio, describing on how detailed measurements can increase or decrease as we change a scale. The idea of fractal dimension can be utilized to solve dimension reduction problems associated with big data. In this work, we demonstrate the usefulness of the fractal dimension as a dimensionality reduction tool using a combination of two scalable algorithms; the box method and the fractal dimensionality reduction algorithm proposed by Traina et al. in 2000. The box method approach allows to partition a data set and compute the fractal dimension. The fractal dimensionality reduction algorithm uses the approach of backward elimination of the features of a given dataset. The underlying concept has been tested using both simulation and real world data set related to gene expression analysis. Finally, we have evaluated the impact from the dimensional reduction in increasing the model accuracy in detecting breast cancer patients using a random forest (RF) algorithm. The RF model with fractal dimension resulted with an accuracy over 75% compared to a model without feature reduction.