How to RETIRE Tabular Data in Favor of Discrete Digital Signal Representation

TL;DR

The paper addresses the challenge of applying deep learning to heterogeneous tabular data by introducing RETIRE, a radar-based multi-dimensional encoding that transforms each tabular instance into a radar chart image for CNN processing. It compares RETIRE against several state-of-the-art MDE methods and XGBoost on 22 benchmark datasets, using ResNet-18 with ImageNet transfer and evaluating transferability, accuracy, and explainability via SHAP analyses. Results show RETIRE often achieves the best balanced accuracy among MDEs and is competitive with XGBoost, with Transfer Learning from ImageNet improving compatibility with CNNs and SHAP explanations aligning with tabular feature importance. The work demonstrates that geometry-based radar representations offer strong generalization, interpretability, and linear encoding-time scalability, suggesting RETIRE as a practical, transferable alternative for tabular-data classification and a step toward broader adoption of MDE techniques.

Abstract

The successes achieved by deep neural networks in computer vision tasks have led in recent years to the emergence of a new research area dubbed Multi-Dimensional Encoding (MDE). Methods belonging to this family aim to transform tabular data into a homogeneous form of discrete digital signals (images) to apply convolutional networks to initially unsuitable problems. Despite the successive emerging works, the pool of multi-dimensional encoding methods is still low, and the scope of research on existing modality encoding techniques is quite limited. To contribute to this area of research, we propose the Radar-based Encoding from Tabular to Image REpresentation (RETIRE), which allows tabular data to be represented as radar graphs, capturing the feature characteristics of each problem instance. RETIRE was compared with a pool of state-of-the-art MDE algorithms as well as with XGBoost in terms of classification accuracy and computational complexity. In addition, an analysis was carried out regarding transferability and explainability to provide more insight into both RETIRE and existing MDE techniques. The results obtained, supported by statistical analysis, confirm the superiority of RETIRE over other established MDE methods.

Figures (6)

• Figure 1: An example of encoding a single sample of a synthetically generated tabular problem with 30 features, into a two-dimensional discrete digital signal using stml, igtd and di techniques.
• Figure 2: Graphical abstract for the proposed method.
• Figure 3: Results of transferability estimation using the TransRate metric. Both bac and TransRate values were averaged for all 22 datasets.
• Figure 4: The results of the experimental evaluation conducted in terms of bac for each of the 22 datasets.
• Figure 5: Explainability of selected data transformation methods applied to the first data fold of the Monkone dataset. The sample numbers in subfigure (c) correspond to the subsequent rows in subfigures (a) and (b). In subfigures (a) and (b), the consecutive columns from the left represent: (i) the original image and its class, (ii) the classifier's decision and the explanation behind it, and (iii) the decision with less support value and its justification.