Federated Learning for Tabular Data using TabNet: A Vehicular Use-Case

TL;DR

This work addresses privacy-preserving classification of vehicular road conditions from time-series data by integrating TabNet with Federated Learning (FL). A feature-extraction step converts time-series data into tabular inputs, which are then trained across multiple edge clients using Flower and FedAvg. The study presents three data sets (Asphalt Regularity, Pavement Type, Asphalt Obstacles) and reports competitive test accuracies up to $93.6\%$, $86.7\%$, and $68.0\%$ respectively, demonstrating the viability of FL for tabular vehicular data and highlighting trade-offs related to dataset size and class balance. The results suggest practical privacy-preserving learning for Intelligent Connected Vehicles (ICVs) and provide a foundation for further exploration of TabNet in FL settings and broader time-series-to-tabular pipelines.

Abstract

In this paper, we show how Federated Learning (FL) can be applied to vehicular use-cases in which we seek to classify obstacles, irregularities and pavement types on roads. Our proposed framework utilizes FL and TabNet, a state-of-the-art neural network for tabular data. We are the first to demonstrate how TabNet can be integrated with FL. Moreover, we achieve a maximum test accuracy of 93.6%. Finally, we reason why FL is a suitable concept for this data set.

Figures (9)

• Figure 1: A FL vehicular scheme: (1) Server sends out initial model, (2) vehicles train model on local data, (3) send back model updates, and (4) server aggregates model updates. This concludes a FL round.
• Figure 2: Example list of features in tabular form christ2018time
• Figure 3: Example of car phone holder used to collect data and directions considered by the accelerometer sensor souza2018asphalt
• Figure 4: Photos that illustrate the four class labels of the Asphalt-Obstacles data set souza2018asphalt
• Figure 5: Performance of asphalt regularity data set.