Keywords

Linear Classifers, Feature Independence, Precision, Accuracy, Recall

Abstract

This paper explores the performance of two fundamental classifcation algorithms. It uses Naive Bayes and K-Nearest Neighbors (KNN), framing it within the context of digit recognition of the MNIST dataset. The MNIST dataset has 70,00 grayscale images of handwritten digits, offering a standard for assessing classifcation models. This paper focuses on key performance metrics such as precision, accuracy, recall, and F1score to examine the effciency of each model. The results reveal that Naive Bayes has moderate accuracy and misclassifcations because of its notion of feature independence. The paper concludes that the KNN model performs better with the optimal k-value of 3, producing the highest accuracy and reducing misclassifcation rates. The comparative analysis helps identify each model’s strengths and limitations and emphasizes the need to explore advanced models in improving and understanding linear classifcation.

Included in

Data Science Commons

Share

COinS