Multi-objective feature selection for warfarin dose prediction. (August 2017)

• Record Type: Journal Article
• Authors: Sohrabi, Mohammad Karim; Tajik, Alireza
• Is Part Of: Computational biology and chemistry. Volume 69(2017)
• Page Start: 126
• Page End: 133
• Publication Date: 2017-08
• Subjects: Warfarin -- Feature selection -- Multi-objective optimization -- Artificial neural networks; Chemistry -- Data processing -- Periodicals; Biology -- Data processing -- Periodicals; Biochemistry -- Data processing; Biology -- Data processing; Molecular biology -- Data processing; Periodicals; Electronic journals; 542.85
• DOI: 10.1016/j.compbiolchem.2017.06.002
• Languages: English
• ISSNs: 1476-9271

Abstract

Graphical abstract: Highlights: The purpose of paper is identification of impressive features using multi-objective optimization algorithms. We propose two new approaches based on NSGA-II and MOPSO to predict warfarin dosage. The prediction of warfarin dose rate is based on Multi-Layer Perceptron in this paper. Multi-objective optimization have more accuracy compared to the classic methods. Abstract: With increasing the application of decision support systems in various fields, using such systems in different aspects of medical science has been growing. Drug's dose prediction is one of the most important issues which can be improved using decision support systems. In this paper, a new multi-objective feature approach has been proposed to support warfarin dose prediction decision. Warfarin is an anticoagulant normally used in the prevention of the formation of clots. This research was conducted on 553 patients during 2013–2015 who were candidates for using warfarin and their INR was in the target range. Features affecting dose was implemented and evaluated, which were clinical and genetic characteristics extracted, and new methods of feature selection based on multi-objective optimization methods such as the Non-dominated Sorting Genetic Algorithm-II (NSGA-II) and Multi-Objective Particle Swarm Optimization (MOPSO) along with the evaluation of artificial neural networks were used. Multi-objective optimization methods have more accuracy and performance compared to the classic methods of feature selection. Furthermore, multi-objective particle swarm optimization algorithm has higher precision than Non-dominated Sorting Genetic Algorithm-II. With a choice of seven features Mean Square Error (MSE), root mean square error (RMSE) and mean absolute error (MAE) were 0.011, 0.1 and 0.109 for MOPSO, respectively.