Immunocomputing-Based Approach for Optimizing the Topologies of LSTM Networks

This paper aims to automatically design optimal LSTM topologies using the clonal selection algorithm (CSA) solve text classification tasks such as sentiment analysis and SMS spam classification. Designing involves determining best configuration of hyperparameters that will give performance. The current state-of-the-art are often designed trial error approaches which incredibly time-consuming require domain experts. Our proposed method, referred CSA-LSTM, is evaluated Large Movie Review Dataset (IMDB). Furthermore, verify robustness discovered by CSA for IMDB dataset, we have used them other datasets, viz. Twitter US Airline Sentiment Spam Collection. Additionally, combined with pre-determined convolutional neural network (CNN) layers achieve same or better results fast training time fewer trainable parameters. For further verification evaluation generalization ability effectiveness approach, it compared four machine learning algorithms widely tasks: (1) random forest (RF), (2) logistic regression (LR), (3) support vector (SVM), (4) multinomial naive Bayes (NB). our experiments show method less expensive, reusable outperform models in literature on three datasets. Through LSTM's topology can be self-determined without any human intervention, making CSA-based a promising approach provide performance given task.