Leveraging Autoencoder and Focal Loss for Imbalanced Data Classification

Class-imbalance is common occurrence in the machine learning and data mining community. Many classification algorithms often fail to learn the minority class information when encountering imbalanced data, resulting in the model being biased towards the majority class. Most importantly, existing learning algorithms mainly concern the class-imbalance distribution, and do not take into account the small number of samples, high feature dimensions, and other issues. To this end, this paper presents a new learning paradigm for solving above problems. To be specific, we first utilize Autoencoder to achieve dimensionality reduction and extract most discriminative features. Next, we explore focal loss to solve the class-imbalance issue and obtain better classification capability. Finally, we tackle the black-box optimization problem via Bayesian Optimization and find a suitable parameters for classification. Extensive experiments performed on three real-world datasets show our method achieves better performance.