Handwritten digit recognition based on multiple basic algorithms

To handle problems of data information recognition in various industries, the issue of handwritten digit recognition is solved in this paper by implementing four machine learning methods for the task of classification. First of all, four different classifiers have been implemented, trained and then the models would be tested. After that, the result is combined to make a final decision. The training data are giving as the MNIST training dataset, and the testing data are using MNIST testing dataset and USPS dataset, the four classifiers which are going to be implemented are Logistic regression, multilayer perceptron neural network, Random Forest model and Support Vector Machine model. Then the program of each algorithm generates the accuracy chart and the normalized confusion matrix to compare their performance in terms of different datasets.