CNN-based approaches for various types of tabular data

Il Do Ha, Pukyong National University

Co-authors: Vu Tuan Anh, Pukyong National University

Abstract: Deep learning (DL) includes various architectures, such as deep neural networks (DNN) and convolutional neural networks (CNN). DL is very powerful and flexible for non-tabular (non-structured) data (e.g. image, text).

However, in tabular data, standard DNNs often do not outperform traditional machine learning (ML) methods such as tree-based models (e.g. random forest, XGBoost). CNNs carry out dimensionality reduction for non-tabular (especially image) data, but may be useful in tabular data too.

In this talk, we present CNN-based approaches for various types of tabular data, which provides an end-to-end learning framework.

The predictive performance of the proposed method is evaluated by comparing it with existing methods using three types of real tabular data, i.e. over-dispersed count data, high-dimension survival data, and time-series data with substantial variability.