(MF)2LS: Memetic framework with memory based fuzzy local search. (15th December 2022)

• Record Type: Journal Article
• Authors: Chatterjee, Bitanu; Ahmed, Shameem; Bhattacharyya, Trinav; Sarkar, Ram
• Is Part Of: Expert systems with applications. Volume 209(2022)
• Publication Date: 2022-12-15
• Subjects: Memetic algorithm -- Optimization -- Fuzzy logic -- Local search -- Feature selection -- Class imbalance; Expert systems (Computer science) -- Periodicals; Systèmes experts (Informatique) -- Périodiques; Electronic journals; 006.33
• DOI: 10.1016/j.eswa.2022.118293
• Languages: English
• ISSNs: 0957-4174

Abstract

Abstract: The Memetic algorithm (MA) breaks down complex optimization problems into smaller sub-parts called memes and operates on them to find optimal solutions. MA uses local search to increase its exploitation capabilities, which makes it a high-performing universal heuristic. This inspires us to work on MA and we attempt to improve the local search ability of the MA which is its core part. Here we propose a new memory-based fuzzy local search method in association with MA with dynamic mutation and a problem specific guided population initialization. This framework is used to solve feature selection and class imbalance problems. Feature selection is a technique used to find the most important subset of features from a high-dimensional dataset to reduce space and computational needs. The class imbalance problem deals with highly imbalanced datasets where the goal is to identify the optimal number of the majority class samples. Otherwise, the classifier becomes biased towards the majority class due to the presence of a huge number of samples and ignores the minority class which might hold the important data. We have named our framework as Memetic Framework with Memory based Fuzzy Local Search ((MF) 2 LS). It has been applied on standard datasets for the respective domains and compared with state-of-the-art methods. For feature selection and class imbalance problems, the proposed method proves to be superior to the state-of-the-art methods. This shows that a good initialization strategy, local search and dynamic mutation contribute significantly to the model's effectiveness. The source code for this method is available on Github . Highlights: A framework using MA is proposed to solve optimization problem. Guided initialization is introduced using the knowledge from datasets. A novel memory based fuzzy local search is incorporated to refine the solutions. The framework is tested using standard publicly available datasets.