This paper introduces and analyzes a new algorithm for linear classification called the voted-perceptron algorithm, which combines Rosenblatt's perceptron algorithm with Helmbold and Warmuth's leave-one-out method. The algorithm is simpler and more efficient than Vapnik's maximal-margin classifier, and can be used effectively in high-dimensional spaces using kernel functions. The algorithm stores information during training and uses it to generate better predictions on test data. It is based on the perceptron algorithm and a transformation of online learning algorithms to batch learning algorithms. The algorithm's performance is close to, but not as good as, maximal-margin classifiers on the same problem, while saving significantly on computation time and programming effort. The paper also shows that the voted-perceptron algorithm can be used with kernel functions to handle high-dimensional data. Experiments on handwritten digit classification show that the algorithm performs well, with performance close to that of maximal-margin classifiers but with faster computation. The algorithm is also compared to other methods, including the adatron and SVMs, and is found to be faster and easier to implement than SVMs. Theoretical analysis of the algorithm shows that it has similar bounds to SVMs for expected error. The paper concludes that the voted-perceptron algorithm is a promising method for linear classification, with potential for further theoretical and practical improvements.This paper introduces and analyzes a new algorithm for linear classification called the voted-perceptron algorithm, which combines Rosenblatt's perceptron algorithm with Helmbold and Warmuth's leave-one-out method. The algorithm is simpler and more efficient than Vapnik's maximal-margin classifier, and can be used effectively in high-dimensional spaces using kernel functions. The algorithm stores information during training and uses it to generate better predictions on test data. It is based on the perceptron algorithm and a transformation of online learning algorithms to batch learning algorithms. The algorithm's performance is close to, but not as good as, maximal-margin classifiers on the same problem, while saving significantly on computation time and programming effort. The paper also shows that the voted-perceptron algorithm can be used with kernel functions to handle high-dimensional data. Experiments on handwritten digit classification show that the algorithm performs well, with performance close to that of maximal-margin classifiers but with faster computation. The algorithm is also compared to other methods, including the adatron and SVMs, and is found to be faster and easier to implement than SVMs. Theoretical analysis of the algorithm shows that it has similar bounds to SVMs for expected error. The paper concludes that the voted-perceptron algorithm is a promising method for linear classification, with potential for further theoretical and practical improvements.