A Russell-Rao Similarity-Based K-Nearest Neighbor Approach for Categorical Data Classification
International Journal of Innovations in Science and Technology · 30.07.2026 · 1745
DOI: 10.33411/ijist/1949 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal of Innovations in Science and Technology, online erschienen am 30.07.2026. Verfasst von Zubair Ahmad und 6 weiteren Autor:innen. Die Arbeit nennt 30 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Machine learning has become a main tool for solving classification problems across several application domains combining vast learning strategies to uncover patterns and produce accurate predictions from the complicated datasets. While many classification techniques have been developed for numerical data comparatively less studies have focused on improving similarity-based methods for the categorical datasets. This research proposes and evaluates three K Nearest Neighbor (KNN) variants developed on the Goodall Gower and Russell Rao similarity measures for categorical data classification. The effectiveness of the built-in models was tested on the Malware Detection Car Evaluation and Mushroom datasets using accuracy precision recall and F1 score. Russell Rao KNN (RRKNN) produced the strongest overall performance obtaining an average accuracy of 90.30% average precision of 90.79% average recall of 90.30% and average F1 score of 90.19%. Goodall KNN (GDKNN) achieved an average accuracy of 77.86% precision of 76.16% recall of 77.86%, and F1 score of 75.53% while Gower KNN (GRKNN) obtained an average accuracy of 76.90% precision of 72.98% recall of 78.52% and F1 score of 75.56%. Across the three datasets RRKNN's accuracy advantage over GDKNN and GRKNN ranged from approximately 0.3 to 39.9 percentage points with the largest gap observed on the Car Evaluation dataset (88.72% vs 57.22% and 48.84%) and the smallest on the Mushroom dataset, where RRKNN and GRKNN differed by only 0.31 points (84.86% vs 84.55%) RRKNN and GRKNN also bounded exactly on Malware Detection (97.33% each).Given the time and resource constraints of this revision, we did not perform a repeated-run statistical significance test. Nevertheless, the observed performance gaps suggest that the difference on the Car Evaluation dataset is unlikely to be explained solely by train-test split variability. In contrast, the differences between RRKNN and GRKNN on the Mushroom and Malware Detection datasets are sufficiently small that the two methods should be considered practically comparable. These results show that Russell Rao KNN consistently matched or outperformed the other two variants across the selected evaluation metrics reflecting its stability and effectiveness for categorical data classification. Overall, the study offers practical guidance for selecting a suitable similarity-based KNN model for data-driven categorical classification tasks and discusses the computational tradeoffs and limitations relevant to real-world deployment.
Zitieren
APA
Ahmad, Z., Anis, M., Ali, N., H, H., Shafiq, S., Abbas, M., & Shafiq, I. (2026). A Russell-Rao Similarity-Based K-Nearest Neighbor Approach for Categorical Data Classification. International Journal of Innovations in Science and Technology, 1745. https://doi.org/10.33411/ijist/1949
Vancouver
Ahmad Z, Anis M, Ali N, H H, Shafiq S, Abbas M, et al. A Russell-Rao Similarity-Based K-Nearest Neighbor Approach for Categorical Data Classification. International Journal of Innovations in Science and Technology. 2026:1745. doi:10.33411/ijist/1949
BibTeX
@article{Ahmad2026,
author = {Ahmad, Zubair and Anis, Muhammad and Ali, Nisar and H, Hasnat and Shafiq, Sana and Abbas, Muhammad and Shafiq, Ibadullah},
title = {A Russell-Rao Similarity-Based K-Nearest Neighbor Approach for Categorical Data Classification},
journal = {International Journal of Innovations in Science and Technology},
year = {2026},
pages = {1745},
doi = {10.33411/ijist/1949}
}
EUVIMED