EFFICIENT QUANTUM COMPUTING DRIVEN AI SOLUTION FOR A LEADING US HEALTHCARE COMPANY

CLIENT

A leading healthcare service provider in the US.

BUSINESS CHALLENGES

The client runs a retail loyalty program aimed at increasing revenue and sales through cross-selling and upselling, while offering membership benefits to customers. The data science team encountered challenges in modeling the program to identify and recommend the loyalty program to the existing and new customers which requires analyzing large datasets with millions of customer records and thousands of attributes to accurately predict customer enrollment probabilities.

Current predictive models employed classical ML techniques and relevant tech stacks for model building and testing the prediction mode. The data science team from the client sought to explore quantum machine learning to streamline modeling, enhance key performance indicators like accuracy,
F1 score, and AUC-ROC, and reduce model build, test, and execution run times.

THE SOLUTION

Mphasis solution was focused on empowering client’s data science team in specific instances of high-dimensional AI-ML problems, and developed a Quantum ML-driven solution pipeline consisting of the following core components :

Quantum Feature Section (QFS) - The QFS component involved optimal search of features with maximum relevancy and minimum redundancy among predictive features and predictive features with target feature.
Quantum Machine Learning (QML) - The QML component addressed the big data and rare event detection complexities which drives computational cost, overfitting, and model performance.
Both QFS and QML pipelines made use of stratified sampling to reduce the data required for feature selection and model training while maintaining model training KPIs.

The tech stack to develop the solution

Azure VM, Azure Data Storage, Azure Quantum Cloud Service (including Azure Quantum Inspired Optimization), D-Wave Hybrid Solvers, Xanadu Pennylane – SDK Quantum Simulators.

BUSINESS BENEFITS

The Quantum ML-driven solution was able to :

Reduce data requirement by 90% without compromising model accuracy

Achieve an AUC-ROC score of 0.7 on the test set

Automate the feature selection and model training process & reduce model building time by 80%