The Speculative Study on Machine Learning Algorithm for an Efficient Classification of Text Document

Document Classification is a problem in information science. The text documents are assigned to one or more classes or categories for classification. Classifying text Documents using various Naive Bayes, Random Forest, Linear and Non-Linear SVM and compared these algorithms on different metrics like accuracy, F1-Score, training and testing time. Based on the analysis, SVM demonstrated that all of the other models when it comes to accuracy. Random Forest accuracy score was also quite good but took considerable time during training phase. From this we can gather the following information. There are three unique or distinct values in the category column which translates to the fact that there are three classes. "Administrative" category seems to be the most frequent category in this column also the count value confirms that there are no missing values in the dataframe now. It can be gathered clearly from the metrics that Linear SVM is clearly the winner in terms of strong accuracy score, training time and testing time. Random Forest and Non linear SVM classifiers' accuracy score is also quite appealing but the high magnitude of their training time is somewhat degrading. It has also been found and demonstrated that if we use lesser number of features (e.g. 5000), selected by applying chi-square test, then the accuracy scores remains the same. The further analyze the training and prediction time of the machine learning algorithms; it will come to further know that it would be greatly improved as well because of much less number of features. The results have been again validated by making use of cross validation. It demonstrated that the accuracy scores that got were equal to the ones got by applying cross validation.