Predictive Repair Management Using a Multi-Head Attention Transformer and Online Learning

2026-06-19

AI Summary

This study proposes an end-to-end deep learning framework to accurately predict vehicle repair durations, aiming to optimize resource allocation and enhance customer satisfaction. The approach uniquely integrates a tabular-data-oriented multi-head attention Transformer with an online learning mechanism, employing an embedding layer to jointly encode categorical and numerical features. Key feature interactions are modeled through the attention mechanism, while a weighted loss function is introduced to mitigate class imbalance. Evaluated on real-world repair records from 2013 to 2020, the model achieves a prediction accuracy of 78%, significantly outperforming feedforward neural networks and random forests. Furthermore, attention weights reveal meaningful associations between vehicle identifiers and repair types, demonstrating the model’s interpretability and practical utility.

Abstract

Accurate prediction of repair duration is an important challenge in product maintenance due to its implications for resource allocation, customer satisfaction, and operational performance. This study aims to develop a deep learning framework to help fleet repair shops accurately categorize repair time given product historical data. The study uses an automobile repair and maintenance dataset and creates an end-to-end predictive framework by employing a multi-head attention network designed for tabular data. The developed framework combines categorical information, transformed through embeddings and attention mechanisms, with numerical historical data to facilitate integration and learning from diverse data features. A weighted loss function is introduced to overcome class imbalance issues in large datasets. Moreover, an online learning strategy is used for continuous incremental model updates to maintain predictive accuracy in evolving operational environments. Our empirical findings demonstrate that the multi-head attention mechanism extracts meaningful interactions between vehicle identifiers and repair types compared to a feed-forward neural network and a random forest model. Also, combining historical maintenance data with an online learning strategy facilitates real-time adjustments to changing patterns and increases the model's predictive performance on new data. The model is tested on real-world repair data spanning 2013 to 2020 and achieves an accuracy of 78%, with attention weight analyses illustrating feature interactions.

Problem

• repair duration prediction
• predictive maintenance
• class imbalance
• online learning
• tabular data

Innovation

• Multi-Head Attention
• Online Learning
• Tabular Data
• Predictive Maintenance
• Class Imbalance