RF Classification for 'Omics' Data

• Activity Number: 652
• Type: Contributed
• Date/Time: Thursday, August 7, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Umashanger Thayasivam*+
• Companies: Rowan University
• Keywords: random forest ; machine learning ; high dimensional ; class lable ; variable importance ; omics data

Abstract

Advance computational approaches are required to extract the complex non-linear trends present in 'omics' data. Recently, application of the machine learning (ML) techniques, that provide computational solutions for managing data complexity. There is a wide range of ML algorithms all of which can perform reasonably good data classification. However, many of these approaches suffer from such deficiencies as inability to estimate the feature importance and poor class identification without significant data pre-processing. The random forest(RF) algorithm is one of the more versatile data classification algorithms in data mining suited for classifying huge amounts of data, with large number of attributes. RF, have shown promise analyzing high dimensional heterogeneous data such as data sets generated by the 'omics' studies. Random Forest algorithm estimates the variable importance for the input features, requires less data pre-processing, and allows for easy interpretation. This talk provides an overview of the RF algorithms and discusses its ability for reasoning with 'omics' data.