Quantum Algorithm Improves Motor Insurance Claim Predictions

Published in Discover Quantum Science, the study analyzes data from Qatar Insurance Company (QIC) to show that leveraging quantum computing properties allows for improved predictive modelling in a highly competitive insurance market. The team employed a hybrid quantum-classical algorithm based on quantum reservoir computing (QRC) to resample original insurance data, aiming to enhance the precision of claim severity predictions for policyholders who have already experienced an accident.

The motivation for this research stems from the intense competition within the motor insurance sector, where even marginal gains in predictive accuracy can translate into significant financial advantages for insurers. In such markets, customers frequently select policies based on the lowest price, meaning that inaccurate risk assessment can lead to substantial financial losses. By improving the ability to predict claim severity, insurance companies can align premiums more closely with potential risks, thereby maintaining financial stability while remaining competitive. The study focuses specifically on claim severity rather than claim frequency or third-party claims, excluding policies without any claims from the analysis to maintain a granular approach to modelling.

The dataset used in the study comprises policyholder and vehicle information from 2018 to 2022, including details on vehicle make, model, body type, channel, agency repair status, sum insured band, insured age group, vehicle age band, insured resident group, policy duration, and policy start dates. To improve prediction accuracy, the researchers filtered the data by vehicle make, model, body type, and channel, predicting claim amounts using the remaining eight features. This granular modelling strategy is a key technique for leveraging big data in insurance, allowing for more precise risk alignment. The claim amounts are expressed in Qatari Riyals, and the analysis accounts for the specific regulatory and demographic context of Qatar, where residency status may influence claim behaviour.

The core of the methodology involves transforming classical insurance claim data into quantum states using angle encoding. In this process, the original data are scaled and normalized to lie within the range of 0 to pi. Each qubit is initialized in the ground state and then rotated around the x or y-axis by the corresponding scaled value using quantum rotation gates. This encoding step is crucial for enabling the subsequent quantum processing, which relies on the probabilistic nature of quantum mechanics to create high-dimensional nonlinear dynamics through entanglement between qubits.

Quantum reservoir computing (QRC) is then applied to resample the feature data. QRC sits at the intersection of recurrent neural networks and quantum computing, utilizing randomised quantum circuits to transform the classical feature matrix into a probabilistic feature matrix. The process involves applying a specified number of reservoir layers, referred to as reservoir depth, to the encoded data. After measurement, a probability distribution is obtained for each row in the original dataset, effectively expanding the feature space from m by n to m by 2^n. This resampling process allows the subsequent machine learning models to operate on a richer set of features derived from the quantum dynamics.

To optimize the quantum circuit, the researchers applied feature selection based on a correlation threshold, reducing the number of required qubits from eight to a range of three to five depending on the vehicle model. This reduction in qubit count is significant because it allows the algorithm to be executed on classical hardware using noise-free simulators, such as IBM’s Qiskit, without the need for large-scale quantum hardware. The study notes that while fault-tolerant quantum computers are expected to become available by 2029, the current approach is immediately applicable for insurance companies using high-performance classical systems, providing a practical advantage in the current noisy intermediate-scale quantum (NISQ) era.

The performance of the quantum-enhanced algorithm was benchmarked against established classical machine learning techniques, including linear regression, Ridge Regression, XGBoost, and CatBoost. The dataset was split into 80% for training and 20% for testing, with the same split applied to all models to ensure a fair comparison. The classical models were extensively optimised via hyperparameter tuning to achieve their best possible performance. The primary metric for evaluation was the root mean squared error (RMSE), which measures the average magnitude of the error between predicted and actual claim amounts. A lower RMSE indicates that the predicted values more closely match the actual data, reflecting higher predictive accuracy.

The results showed that the hybrid quantum model consistently achieved lower RMSE scores than all classical models across the various vehicle models tested. For some vehicle models, the difference in RMSE performance between the classical and quantum approaches exceeded 20%. This improvement in accuracy suggests that the quantum-based resampling process provides a more effective representation of the underlying risk factors in the insurance data. The study also evaluated the algorithm under less granular settings, filtering only by vehicle model and using ten features, and found that the quantum-enhanced algorithm still yielded lower RMSE values compared to its classical counterparts.

An important aspect of the quantum reservoir computing process is the randomisation of the quantum circuit, which can lead to variations in the probabilistic feature matrix and, consequently, in the prediction performance. The researchers investigated the impact of this randomisation by running multiple iterations with different random parameters. They found that the minimum RMSE and the mean of the three smallest RMSE values stabilised once the number of iterations became sufficiently large. While increasing the number of iterations can further improve accuracy, it also increases computational cost. The study suggests that choosing around 10 iterations provides a balance between accuracy and computational efficiency, offering better performance than classical models without excessive resource usage.

The study highlights that the integration of gradient boosting methods with quantum reservoir computing within a hybrid algorithm has not been previously explored in the context of motor insurance claim analysis. This novelty addresses a gap in the literature where quantum machine learning approaches had not yet demonstrated a predictive advantage over boosting methods in this specific domain. The findings suggest that quantum-enhanced algorithms can offer a short-term competitive advantage to early adopters in the insurance industry, potentially leading to more accurate pricing and risk management. However, the widespread adoption of such methods is expected to intensify market competitiveness over the long term as more insurers incorporate these advanced analytical tools.

Limitations of the study include the reliance on simulated quantum hardware rather than physical quantum devices, although the authors argue that the method does not require a large number of qubits, making it suitable for current classical hardware. The dataset is specific to Qatar Insurance Company, and while the authors suggest that the approach should produce similar accuracy in other insurance companies’ datasets in the same region, generalizability to different regulatory and demographic contexts has not been tested. Additionally, the study focuses on claim severity prediction and does not address claim frequency or fraud detection, which are other critical aspects of insurance analytics. The confidentiality of the insurance data prevents public sharing of the dataset, limiting independent replication of the results.

As the insurance industry continues to face evolving risks, including climate change and cyber threats, the need for new analytical methods becomes increasingly apparent. The study demonstrates that quantum machine learning offers a viable path for improving predictive accuracy in motor insurance claim analysis. By leveraging the properties of quantum computing, insurers can potentially enhance their risk assessment capabilities, leading to more stable financial performance in a competitive market. The immediate applicability of the algorithm on classical hardware positions it as a practical tool for current industry needs, while also serving as a foundation for future applications as fault-tolerant quantum computers become available.

Subject of Research: Motor insurance data analysis by quantum machine learning

Article Title: Motor insurance data analysis by quantum machine learning

Article References: Tamturk, M., Ginossar, E., Carenzo, M., & Perret, C. (2026). Motor insurance data analysis by quantum machine learning. Discover Quantum Science, 2(1), Article 3. https://doi.org/10.1007/s44464-026-00007-x

Image Credits: AI Generated

DOI: 10.1007/s44464-026-00007-x

Keywords: Motor, insurance, data, analysis, quantum, machine, learning, scientific research