Interval Cost Feature Selection Using Multi-objective PSO and Linear Interval Programming

Interval cost feature selection problems (ICFS) are popular in real-world. However, since the optimized objectives not only are multiple but also contain interval coefficients, there have been few solving methods. This paper first transforms the ICFS into a multi-objective one with exact coefficients by the linear interval programming. Second, by combining a multi-objective particle swarm algorithm (which has a good performance in exploration) with a powerful problem-specific local search (which is good at exploitation), we propose a memetic multi-objective feature selection algorithm (MMFS-PSO). Finally, experimental results confirmed the advantages of our method.