Purpose: The purpose of this research is to evaluate six classification algorithms for handwritten digit recognition. The study aims to assess their accuracy and effectiveness using established and custom datasets. It provides valuable insights for algorithm selection, guiding researchers and practitioners in handwriting recognition. Methodology: The study utilises a comparative analysis approach to evaluate the performance of the classification algorithms. Two datasets are employed: the established load_digits dataset and a custom dataset of handwritten digits. Confusion matrices are used to measure accuracy, and the algorithms are assessed based on their performance on the respective datasets. Findings: The findings of this study reveal that KNN exhibits the highest accuracy in recognising handwritten digits, achieving remarkable accuracy scores of 0.99 on the custom dataset and 0.95 on the load_digits dataset. SVM closely follows with accuracies of 0.98 on the custom dataset and 0.96 on the load_digits dataset. Additionally, SGD demonstrates strong performance with accuracies of 0.96 and 0.90 on the respective datasets. GNB, DT, and RF show comparable performance levels, with DT recording the lowest accuracy. Originality/Value: This research provides valuable insights and originality in evaluating multiple classification algorithms for handwritten digit recognition. It aids researchers and practitioners in algorithm selection for optimal performance, utilising both established and custom datasets. The findings contribute to the existing knowledge base in handwriting recognition and suggest future directions for algorithm refinement and exploration of ensemble methods or deep learning approaches.

SVM, KNN,Gaussian Naive Bayes, Random forests, Stochastic Gradient Descent