HSDLM: A Hybrid Sampling With Deep Learning Method for Imbalanced Data Classification

Imbalanced data presents many difficulties, as the majority of learners will be prejudice against the majority class, and in severe cases, may fully disregard the minority class. Over the last few decades, class inequality has been extensively researched using traditional machine learning techniques. However, there is relatively little analytical research in the field of deep learning with class inequality. In this article, the authors classify the imbalanced data with the combination of both sampling method and deep learning method. They propose a novel sampling-based deep learning method (HSDLM) to address the class imbalance problem. They preprocess the data with label encoding and remove the noisy data with the under-sampling technique edited nearest neighbor (ENN) algorithm. They also balance the data using the over-sampling technique SMOTE and apply parallelly three types of long short-term memory networks, which is a deep learning classifier. The experimental findings indicate that HSDLM is a promising and fruitful solution to working with strongly imbalanced datasets.