Analysis of Performance Metrics for Classification of Punjabi Poetry using Machine Learning Techniques

The poetic text has always been a challenge from the computational linguistics perspective. Literature classification has always been a tedious task in itself owing to so much diversity we have. We have so much information associated with a poem that it can be classified on basis of its writing schemes, on basis of poets, on the grounds of emotions, on basis of an era, etc. In this work, we developed a dataset of 401 poems in total and they were classified into four categories related to different genres. Considering the lexical features of poems, we performed several pre-processing stages to segregate the useful information in form of tokens from poems. These tokens were modeled using Latent Dirichlet Allocation and evaluated using the Term Frequency (TF) and Term Frequency-Inverse Document Frequency (TF- IDF) weighting techniques. SVM, Random Forest, Naive Bayes, and Logistic Regression experimented with lexical features in the poem and it was found that SVM proved to be the best- performing algorithm with the highest accuracy of 82.17%.