Handwritten Digit Recognition using Ensemble learning techniques: A Comparative performance Analysis

The detection of some digits written by humans may seem very easy; however, in reality, it is a very challenging and tedious task to extract text from real images because of the wide variety of font sizes, shapes, textures and backgrounds, etc. This study aimed to recognize handwritten digit images by extracting the features from datasets. Then theses learned characteristics are subsequently submitted to the classifier to improve the overall performance. The Bagging and boosting techniques of ensemble learning were performed by integrating different base models. Each single base model corrected the errors of the previous model. Six bagging and boosting ensemble learning algorithms were used to recognize the handwritten digit images. Therefore, with 98% of accuracy, the extreme gradient boosting machine outperformed the other methods in terms of overall performance. A detailed analysis of the allocation of data among the base learners and their overall performance, including the model correctness, computation time, and error rate, was evaluated and projected.