An enhance multimodal multiobjective optimization genetic algorithm with special crowding distance for pulmonary hypertension feature selection

• Published in: Computers in biology and medicine Vol. 146; p. 105536
• Main Authors: Wang, Mingjing, Li, Xiaoping, Chen, Long
• Format: Journal Article
• Language: English
• Published: United States Elsevier Ltd 01.07.2022; Elsevier Limited
• Subjects: Accuracy; Algorithms; Blood gas analysis; Classification; Crossovers; Crowding; Crowding distance calculation; Feature selection; Gas analysis; Gene mapping; Genetic algorithm; Genetic algorithms; Genetic crosses; Hypertension; Internal Medicine; Methods; Multimodal multiobjective optimization; Multiobjective optimization; Multiple objective analysis; Optimization; Other; Pareto optimization; Pareto optimum; Pulmonary arteries; Pulmonary hypertension; Statistical analysis; Feature selection; Multimodal multiobjective optimization; Crowding distance calculation; Genetic algorithm; Multiobjective optimization; Pulmonary hypertension
• ISSN: 0010-4825; 1879-0534; 1879-0534
• DOI: 10.1016/j.compbiomed.2022.105536

Multiobjective optimization assumes a one-to-one mapping between decisions and objective space, however, this is not always the case. When many variables have the same or equivalent objective value, a multimodal multiobjective issue develops in which more than one Pareto Set (PS) maps to the same Pareto Front (PF). Evolutionary computing research into multimodal multiobjective optimization issues has increased (MMOPs). This paper proposed an enhanced multimodal multiobjective genetic algorithm to crack MMOPs using a special crowding distance calculation (ESNSGA-II). This special crowding distance calculation can consider the diversity of the decision space while paying attention to the diversity of the object space. Then, a unique crossover mechanism is established by combining the simulated binary crossover (SBX) method with the capacity of Pareto solutions to generate offspring solutions. The balance between convergence and diversity in both decision space and object space can be guaranteed synchronously, and PS distribution and PF accuracy may both be enhanced. The proposed ESNSGA-II uses the CEC2020 benchmarks MMF1-MMF8 to assess its properties. Comparing the ESNSGA-II to other recently established multimodal multiobjective evolutionary techniques demonstrates that it is capable of efficiently searching numerous PSs of MMOPs. Finally, the suggested ESNSGA-II is used to address a real MMOP problem of pulmonary hypertension detection via arterial blood gas analysis. The statistical analysis reveals that the suggested ESNSGA-II algorithm outperforms other algorithms on this MMOP, and so may be considered a possible tool for pulmonary hypertension. •The ESNSGA-II is presented for MMOPs.•Performance of the ESNSGA-II is enhanced by special crowding distance.•The balance between convergence and diversity of decision and object space can be guaranteed.•The efficacy of ESNSGA-II is shown on benchmarks and is much superior to other methods.•ESNSGA-II may be treated as tool for assist PH diagnosis.