An adaptive inertia weight teaching-learning-based optimization algorithm and its applications

• Published in: Applied Mathematical Modelling Vol. 77; pp. 309 - 326
• Main Authors: Shukla, Alok Kumar, Singh, Pradeep, Vardhan, Manu
• Format: Journal Article
• Language: English
• Published: New York Elsevier Inc 01.01.2020; Elsevier BV
• Subjects: Accuracy; Adaptive algorithms; Benchmark; Classification; Datasets; Gene expression; Gene selection; Heuristic methods; Inertia; Inertia weight; Machine learning; Metaheuristic; Optimization; Optimization algorithms; Probability distribution functions; Teaching learning-based optimization; Weight; Benchmark; Inertia weight; Metaheuristic; Gene selection; Teaching learning-based optimization; Classification
• ISSN: 0307-904X; 1088-8691; 0307-904X
• DOI: 10.1016/j.apm.2019.07.046

•The logistic-map is used to generate uniformly distributed individuals to enhance the original population quality.•A new inertia weight strategy is used in teacher phase of TLBO to increase the learning capacity of the learners.•The random topological order is adopted by using inertia weight, so neighborhood learner improves their search capability.•Compare with three variants of inertia weight, such as time-varying, adaptive, and constant inertia weight.•Adaptive teaching-learning based optimization is also used for gene selection by proposing new updating mechanisms. This paper presents an effective metaheuristic algorithm called teaching learning-based optimization which is widely applied to solve the various real-world optimization problems. However, teaching learning-based optimization is rapidly trapped into local optima. To handle this kind of problem, we proposed an improved teaching learning-based optimization algorithm using adaptive exponential distribution inertia weight and altering the position-updating equation. In addition, the logistic map is applied to generate a uniformly distributed population to enhance the quality of the initial populations. The performance of the proposed method is evaluated on a suite of benchmark functions with different characteristics. The efficiency of the proposed technique is also evaluated on six gene expression datasets with the help of three classifiers. The experimental result demonstrates that the proposed method is comparatively useful in adapting the inertia weight in comparison to the existing inertia weight strategies with regards to the quality of solutions, convergence rate along with classification accuracy. In particular gene selection, the proposed method has achieved up to 98% classification accuracy for three out of six datasets with optimal gene subsets for all six datasets and maximum accuracy is achieved as 100% in small round blue-cell tumor dataset.