A New Classification Technique: Random Weighted LSTM (RWL)
Abu Sadat Al Rafi, Abu Sadat Al Rafi, Tauhidur Rahman, Tauhidur Rahman, Abdur Rahman Al Abir, Abdur Rahman Al Abir, Tanvir Ahmed Rajib, Tanvir Ahmed Rajib, Muhayminul Islam, Muhayminul Islam, Md. Saddam Hossain Mukta, Md. Saddam Hossain Mukta · 2020 IEEE Region 10 Symposium (TENSYMP) · 2020
Due to the unprecedented growth of digital devices and hand held smartphones, users generate several quintillion of data everyday by using social media, blog, youtube, etc. With the advancement of machine learning techniques, we classify these data automatically for critical decision making process. Majority of these classification algorithms are linear in nature and these algorithms show weak performance in predicting class labels when the attributes are complex. For example, human behavior, preference, personality, etc. have numerous non-linear properties and difficult to predict in real life by using these traditional machine learning algorithms. In this paper, we propose a novel non-linear technique based on Long short-term memory (LSTM) architecture. Studies show that Recurrent Neural Network (RNN) and LSTM based models usually predict time and sequential models better than that of other models. We significantly change the operational mechanism of LSTM and achieve outstanding performance in predicting classification problems. We run our algorithm over six different datasets: Iris, Pima Indian, Breast Cancer, Blood Transfusion, StackOverflow, and Banknote Authentication. We compare the performance of our algorithm with other traditional classifiers. Our classifier generally outperforms conventional linear and non-linear classifiers.