sis 18: e1

Research Article

An Improved Weighted Base Classification for Optimum Weighted Nearest Neighbor Classifiers

Download140 downloads
  • @ARTICLE{10.4108/eai.13-7-2018.163339,
        author={Muhammad Abbas and Kamran Ali Memon and Noor ul Ain and Ekang Francis Ajebesone and Muhammad Usaid and Zulfiqar Ali Bhutto},
        title={An Improved Weighted Base Classification for Optimum Weighted Nearest Neighbor Classifiers},
        journal={EAI Endorsed Transactions on Scalable Information Systems: Online First},
        volume={},
        number={},
        publisher={EAI},
        journal_a={SIS},
        year={2020},
        month={2},
        keywords={Classification, k-Nearest Neighbor (kNN), Logistic Regression, Decision Trees, Cross-Validation, Machine-Learning (ML), SVM, random forest, improved version of k-nearest neighbor (IVkNN), and Python},
        doi={10.4108/eai.13-7-2018.163339}
    }
    
  • Muhammad Abbas
    Kamran Ali Memon
    Noor ul Ain
    Ekang Francis Ajebesone
    Muhammad Usaid
    Zulfiqar Ali Bhutto
    Year: 2020
    An Improved Weighted Base Classification for Optimum Weighted Nearest Neighbor Classifiers
    SIS
    EAI
    DOI: 10.4108/eai.13-7-2018.163339
Muhammad Abbas1, Kamran Ali Memon2,*, Noor ul Ain3, Ekang Francis Ajebesone3, Muhammad Usaid4, Zulfiqar Ali Bhutto5
  • 1: School of Computer Science, Beijing University of Posts and Telecommunications Beijing, China
  • 2: School of Electronic Engineering, Beijing University of Posts & Telecommunications, Beijing, China
  • 3: School of Information & Communication Engineering, Beijing University of Posts & Telecommunications, Beijing, China
  • 4: Department of Electrical Engineering, Mehran University of Engineering & Technology, Jamshoro, Pakistan
  • 5: Dawood University of Engineering and Technology, Karachi, Pakistan
*Contact email: Ali.kamran77@gmail.com

Abstract

Existing classification studies use two non-parametric classifiers- k-nearest neighbours (kNN) and decision trees, and one parametric classifier-logistic regression, generating high accuracies. Previous research work has compared the results of these classifiers with training patterns of different sizes to study alcohol tests. In this paper, the Improved Version of the kNN (IVkNN) algorithm is presented which overcomes the limitation of the conventional kNN algorithm to classify wine quality. The proposed method typically identifies the same number of nearest neighbours for each test example. Results indicate a higher Overall Accuracy (OA) that oscillates between 67% and 76%. Among the three classifiers, the least sensitive to the training sample size was the kNN and produced the unrivalled OA, followed by sequential decision trees and logistic regression. Based on the sample size, the proposed IVkNN model presented 80% accuracy and 0.375 root mean square error (RMSE).