Data Classification with k-fold Cross Validation and Holdout Accuracy Estimation Methods with 5 Different Machine Learning Techniques
Kaushika Pal, Biraj V. Patel
Abstract
Classification of documents is measured in terms of accuracy by comparing the actual labels with predicted labels for the classes. There are many machine learning techniques, which can be used to build a classifier, it is almost difficult to manually predict which technique should be used for classification, especially when we are working on Indic languages and there is no reliable method, which can give good results. Such area still need to be explored by processing documents with natural language processing, then applying machine learning techniques to build the classifier. This research article has collected 154 Hindi Poetries from web; processed it using NLP techniques, extracted features to build classifier with 5 Machine Learning methods namely support vector machine, naïve bayes, decision tree algorithm, random forest and k-nearest neighbors. The results were examined in accuracy with holdout accuracy estimator and k-fold cross validation accuracy estimator to check the reliability of the methods. The results of this research will help to select the best document classifier for Hindi Poetries and increase accuracy by working on various feature extraction techniques and NLP techniques. The result of the experiment shows that the results of SVM, NB and random forest methods are better as compared to DTT and K-NN for used data set available in this experiment.