Runtime Optimization of Widrow-Haff Classification Algorithm Using Proper Learning Samples

Link to the paper:

• https://www.researchgate.net/publication/319015521_Runtime_Optimization_of_Widrow-Hoff_Classification_Algorithm_Using_Proper_Learning_Samples
• Or
• https://civilica.com/doc/668873/

Please cite:

• Mirinezhad, S. Younes, et al. "Runtime Optimization of Widrow-Haff Classification Algorithm Using Proper Learning Samples." 4th National Conference on Information Technology, Computer & Telecommunication. Mashhad, Iran. 2017.
• Or
• Dezfoulian, Mir-Hossein and Mirinezhad, S. Yunes and Mousavi, S. M. Hussein and Shafeii Mosleh, Mehrdad,1396,Runtime Optimization of Widrow-Haff Classification Algorithm Using Proper Learning Samples,4th National Conference on Information Technology?Computer & Telecommunication ,Mashhad,https://civilica.com/doc/668873

Overview

This repository provides an implementation of the methodology described in the paper: "Runtime Optimization of Widrow-Hoff Classification Algorithm Using Proper Learning Samples." The paper introduces a novel combination of the Widrow-Hoff classification algorithm and the Multi-Class Instance Selection (MCIS) method to optimize the classification runtime and accuracy.

---

Key Contributions of the Paper

1. Widrow-Hoff Algorithm:

   • A fast and efficient linear classification algorithm.
   • Sensitive to outliers and noisy data, which can impact its performance.

2. MCIS (Multi-Class Instance Selection):

   • Pre-selects the most relevant data samples for classification.
   • Reduces the dataset size by removing noisy or outlier samples.
   • Improves runtime efficiency and accuracy when used with Widrow-Hoff.

3. Combination:

   • Integrating MCIS with Widrow-Hoff significantly improves runtime performance while maintaining or enhancing classification accuracy.

Methodology

1. MCIS Algorithm:

   • Clusters positive class samples using K-Means.
   • Filters outlier samples from the negative class based on their distance from cluster centers.
   • The filtered dataset is passed to the classifier.

2. Widrow-Hoff Algorithm:

   • A linear classifier that iteratively updates weights using the Widrow-Hoff learning rule: ( w = w + \eta (y - \hat{y})x )
   • Trained on the filtered dataset from MCIS.