To successfully classify poems as science and non-science, science poems were defined based on determined criteria and poems were collected and scraped to build the dataset. The used dataset contained 468 science poems collected online from Collecting Realities, Poets.org and Consilience Journal. Six classification models were developed: DBSCAN and KMeans clustering methods, topic modelling with LDA, transformer-based BERT model and neural networks BiLSTM and CNN+BiLSTM models.