NIC: An innovative supervised machine learning computational framework having network‐based interactional connections. (15th June 2021)

• Record Type: Journal Article
• Authors: Mirrashid, Alireza; Shirazi, Ali‐Asghar Beheshti
• Other Names: Ventura Sebastian guestEditor.; Soda Paolo guestEditor.; González Alejandro Rodríguez guestEditor.
• Is Part Of: Computational intelligence. Volume 37:Number 4(2021)
• Page Start: 1916
• Page End: 1943
• Publication Date: 2021-06-15
• Subjects: interactional connection -- neural network -- pattern recognition -- soft computing; Artificial intelligence -- Periodicals; Computational linguistics -- Periodicals; 006.3
• DOI: 10.1111/coin.12472
• Languages: English
• ISSNs: 0824-7935

Abstract

Abstract: Models based on computational intelligence have many applications in various problems. Such systems are generally set up and designed based on a collection of data and information. In some real problems, the implementation of experimental studies or doing tests are costly and time‐consuming. Therefore, a model which requires fewer data than the existing soft computing methods can be useful and applicable. In this article, a network‐based interactional connection system is proposed as a new supervised machine learning computational framework for problems with small data. This model, which is inspired by the connections between neurons of the brain, utilizes the series and parallel structures with interactional connections to determine the best estimation. The proposed approach uses less unknown parameters than existing models and gives a suitable response in a few steps. The learning process starts with one generation and continues until an acceptable prediction is found, and the coefficients are determined. However, it may have more generations in a sequential format for better prediction and providing more accurate answers. To evaluate the performance of the proposed computational system, three engineering problems were investigated as a numerical study. The results also compared with the predicted values of four well‐known techniques.