The computational analysis of the Bangla poems is a challenging task due to the diverse linguistic, stylistic, and semantic features of the Bangla language. In this work, we prepared a dataset of 1311 Bangla poems of two separate categories: Love and Miscellaneous poem, which contain 500 and 811 poems respectively. We used word or semantic-based features to classify Bangla poems using the TF-IDF feature techniques. We used Logistic Regression, Naïve Bayes (NB), and Support Vector Machine (SVM) models for classification through machine learning, and we used Bayesian optimization techniques for hyperparameters tuning of these three models. We also used LSTM, CNN, and transformer models for this research. For the performance evaluation of the classification models, we used four evaluation metrics of precision, recall, F1-score, and accuracy. We also used the ROC-AUC curve to distinguish between all the machine learning and deep learning models. The experimental results expressed that, the transformer model achieved the highest accuracy compared to all the typical machine learning and deep learning models with an accuracy of 87%.