Design of belief theoretical methods for improved classification of breast tumors

Abstract

Belief theory or Dempster Shafer theory involves the use of Shafer s model and combination rules. Shafer s model extends the general probabilistic model to deal with newlinethe ignorance condition, avoiding the assignment of lack of belief about a hypothesis newlineto its negation as done in the probabilistic model. This makes it very relevant for use newlinein classification scenarios as seen in the design of the evidential k-nearest neighbours newline(EKNN) algorithm improving upon the conventional k-nearest neighbours (KNN) newlinealgorithm. The combination rules provided by the belief theoretical framework en- newlineables the effective fusion of information from multiple sources. Several combination newlinerules including Dempster s rule, Dubois and Prade rule, disjunctive rule, and so on, are newlineavailable. The difference between them is in the way they handle conflict between newlinethe sources under consideration, which impacts the combination rule choice for the newlinechosen scenario. These tools offered by belief theory can improve upon existing newlineclassification algorithms and aid in the design of highly precise decision making newlinesystems. This makes it particularly suitable for use in biomedical applications, where newlinethe margin for error is small and the consequences of making the wrong decision can newlinebe severe. Development of automated systems for breast cancer diagnosis has been a newlineresearch problem which has received wide attention of late. This research analyses newlinethe application of Dempster Shafer theory to classification problems and looks at newlinedeveloping and improving classification methodologies with a focus on breast cancer newlinemalignancy classification. newline