Evaluation of supervised models for classification models using EMNIST extended dataset
The wide availability of images has prompted the identification of the most powerful and accurate algorithms for the task. We would like to understand which algorithm out of a selected group(RandomForest, Neural Network(NN) and Convolutional Neural Network(CNN)) is the most effective in predicting items from our input data-set. For this task the extended MNIST image database (EMNIST) is adequate. Once the best model is discovered we also want to understand what effect noisy/altered images have on the model’s ability to predict correctly.
The dataset used during this project consists on a variant of the NIST datdaset, called extended MNIST (EMNIST). IT consists on set of datasets that consitutes a more challenging classification task involving letters an ddigits, sharing the same image structure and parameters as the original MNIST task, hance its compatibility with the already existing classifiers and systems. The dataset contains handwritten digits and characters collected from 500 writers. The database intended to provide a more complex optical character recognition, therefore presents the data in five separate organizations.