Feature Dimensionality Reduction for Mammographic Report Classification

Abstract

The amount and the variety of available medical data coming from multiple and heterogeneous sources can inhibit analysis, manual interpretation and use of simple data management applications. In the healthcare domain, the development of techniques for enabling health data management, analysis, mining and recognition has become worldwide important. In this paper, the use of the most-known dimensionality reduction techniques on a dataset composed of real mammographic reports is presented. Techniques such as LSI, PCA, and SVD decomposition have been applied to the extracted TF-IDF matrix using less attributes than the original unprocessed matrix and obtaining comparable results. Due to their reliability, LSI and PCA techniques can be efficiently used, increasing any computation feasibility on reduced feature data.