A fuzzy multi-objective hybrid TLBO–PSO approach to select the associated genes with breast cancer

• Taking into account two conflicting goals: the number of genes and the classification accuracy.
• A new TLBO binarization formula is proposed.
• A novel hybrid TLBO–PSO method is proposed.
• A new mutated version of the particle swarm optimization is proposed.
• By using the data folding technique and repeating the assessment of each subset of selected genes in this manuscript, the randomness of the results is reduced and also the robustness of the proposed technique is increased.

When the genes associated with breast cancer are mutated, they may not function normally and breast cancer risk increases. Therefore the method that among huge number of unrelated genes identifies the genes associated with breast cancer is an efficient method for diagnosis of breast cancer before the progression of the disease. In this paper, a new hybrid algorithm is proposed to identify the most relevant genes involved in breast cancer development. A combination of the teaching learning-based optimization (TLBO) algorithm and the proposed mutated fuzzy adaptive particle swarm optimization (PSO) algorithm is employed to find the smallest subset of genes involved in breast cancer with the highest amount of classification accuracy, sensitivity and specificity. Due to the presence of the two conflicting goals, i.e. minimization of the number of selected genes and maximization of the classification performance, the optimization problem is represented in a multi-objective form and solved using the Pareto technique. The obtained results show that the proposed technique is able to achieve the accuracy of 91.88%, the sensitivity of 90.55% and the specificity of 93.33% in the breast cancer microarray data by selecting 195 genes.