A Quantum Support Vector Regression Framework for Missing Data Imputation in Ship Performance Data

Daniel Maroto Sanchez, Christian Velasco-Gallego, Miriam Bastante Chichon, Jaime Álvarez-Benayas, Tsoulakos Nikolaos, J. A. Bravo-MontesarXiv 2026 · arXiv

TL;DR

This paper introduces the first application of Quantum Support Vector Regression (QSVR) for imputing missing sensor data in the maritime industry. By leveraging quantum feature maps to model complex relationships in ship performance telemetry, the researchers demonstrate that quantum-enhanced models can match or slightly outperform state-of-the-art classical methods, maintaining high accuracy even when 90% of the data is missing.

What Problem Does This Paper Address?

The "Internet of Ships" has turned modern vessels into floating data centers, generating vast amounts of telemetry. However, this data is frequently plagued by missing values due to sensor malfunctions, network instability, or human error. These gaps compromise the reliability of critical downstream applications like route optimization, fuel consumption prediction, and predictive maintenance.

While classical machine learning has been the standard for "filling in the blanks" (data imputation), the maritime domain has yet to explore the potential of Quantum Machine Learning (QML). The authors argue that quantum kernel methods might provide superior representations for the complex, non-linear relationships found in high-frequency ship sensor data by mapping that data into high-dimensional Hilbert spaces.

Methods

The researchers developed a framework comparing classical Support Vector Regression (SVR) against its quantum counterpart, QSVR. The core difference lies in the kernel function—the mathematical tool used to measure similarity between data points.

Figure 1. Overview of the experimental workflow used to compare classical SVR and QSVR models for missing-data imputation.

Key Components:

  • Classical Baselines: SVR using standard kernels like Linear, Polynomial, and Radial Basis Function (RBF).
  • Quantum Feature Maps: The study tested two main types of quantum embeddings to translate classical data into quantum states:
  • Angle Encoding: Maps features to single-qubit rotations.
  • ZZFeatureMap: A more complex map that introduces non-linear feature couplings and entanglement to capture correlations between variables.
  • Entanglement Topologies: They explored different ways to connect qubits (Linear, Circular, and Full) to find the best way to represent feature interactions.
  • Hybrid Workflow: While the kernel is estimated using quantum circuits (via statevector emulation), the optimization and prediction stages remain classical, creating a hybrid quantum-classical pipeline.
Figure 2. Common modelling framework for classical SVR and quantum-kernel-based QSVR.

Datasets

The study utilized a high-frequency operational dataset from a Panamax bulk carrier.

  • Scope: 397 operational days (Jan 2024 – Jan 2025).
  • Resolution: 1-minute intervals, totaling approximately 571,680 observations per variable.
  • Features: 37 variables across six categories, including engine performance (RPM, fuel load), propulsion (torque, thrust), navigation (speed, heading), and environmental conditions (wave height, wind speed).
  • Target: The models were tasked with imputing the main-engine fuel-flow rate.

Key Insights and Ideas

  • Quantum Viability: The study proves that quantum feature maps can effectively capture the physical laws governing marine propulsion. The ZZFeatureMap was particularly effective, likely because its structure naturally accounts for the inter-dependencies between engine parameters and environmental resistance.
  • Extreme Scarcity Resilience: One of the most surprising findings was the robustness of these models. Even with a "Missing Completely At Random" (MCAR) rate of 90%, the best models maintained an $R^2$ above 0.99.
  • Dimensionality Diminishing Returns: In the scalability analysis, the authors found that adding more than 14 features (and thus 14 qubits) yielded limited accuracy gains but caused an exponential spike in the computational resources required for quantum emulation.
  • The RBF Benchmark: While quantum models won on pure accuracy, the classical RBF kernel proved to be an incredibly strong and efficient baseline, highlighting that for current practical applications, classical methods are still very hard to beat.

Key Results

  • Top Performer: The ZZFeatureMap_1 configuration (using linear entanglement and one repetition) achieved the highest accuracy.
  • Accuracy: In a representative test (60% dataset, 90% missing data, 14 features), ZZFeatureMap_1 reached an $R^2$ of 0.9912 and the lowest RMSE of 19.51 kg/h.
  • Classical Comparison: The RBF kernel was a close second with an $R^2$ of 0.9896.
  • Efficiency Gap: There is a massive disparity in runtime. The classical RBF model took only 1.78 seconds, whereas the emulated QSVR required 1,439.76 seconds. This reflects the high cost of simulating quantum circuits on classical hardware.

Limitations and Future Work

  • Emulation vs. Hardware: The results are based on "exact statevector emulation" (simulating a perfect quantum computer on a classical CPU). Real-world quantum hardware introduces "noise" that might degrade these results.
  • Missingness Mechanisms: The study focused on data missing "completely at random." Future research needs to address more realistic scenarios where data might be missing due to specific conditions (e.g., sensors failing only at high temperatures).
  • Computational Scaling: The exponential memory requirements for simulating more qubits currently limit the number of features that can be explored.
  • Next Steps: The authors suggest investigating hybrid quantum-classical architectures and testing the framework on actual Quantum Processing Units (QPUs) to assess true computational speedups.

Summary updated

Abstract

The Internet of Ships enables vessels to generate vast amounts of data, yet missing values remain a persistent issue. To the best of the authors' knowledge, existing imputation techniques in this domain rely exclusively on classical machine learning. This study investigates quantum support vector regression (QSVR) for imputing missing values in ship performance data. Classical support vector regression and QSVR models based on different quantum feature maps are compared for estimating marine fuel consumption under varying levels of data availability. The case study uses real-world data from a Panamax bulk carrier, covering 37 variables across six categories. Three experiments are conducted: exploratory analysis with three input features, scalability analysis with increasing dimensionality (3 to 18 features), and comparative evaluation under a selected representative configuration. Across all scenarios, both classical and quantum approaches captured the relationship between the inputs and the main-engine fuel-flow rate, maintaining strong imputation performance even under severe data scarcity. ZZFeatureMap_1 achieved the highest average R2 (0.991 in the performance comparison conducted with 60% of the dataset, 90% missing data, and 14 input features, for instance) and the lowest aggregate imputation errors, while the radial basis function kernel was the strongest classical baseline, delivering highly competitive results.
BibTeX
@misc{sanchez2026quantum,
  author = {Daniel Maroto Sanchez and Christian Velasco-Gallego and Miriam Bastante Chichon and Jaime Álvarez-Benayas and Tsoulakos Nikolaos and J. A. Bravo-Montes},
  title = {{A Quantum Support Vector Regression Framework for Missing Data Imputation in Ship Performance Data}},
  year = {2026},
  eprint = {2611.22635},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2611.22635},
}
Figure