Robust evaluation of classical and quantum machine learning under noise, imbalance, feature reduction and...

Area of Science:

• Computer Science
• Quantum Computing
• Artificial Intelligence

Background:

• Real-world data often presents challenges like noise, imbalance, and high dimensionality, impacting traditional machine learning (ML) model performance.
• Quantum machine learning (QML) offers potential advantages for complex classification tasks by utilizing quantum computation.
• Evaluating QML models under realistic data conditions is crucial for practical deployment.

Purpose of the Study:

• To conduct an extensive experimental comparison of traditional supervised ML classifiers and QML classifiers.
• To assess model performance on diverse datasets with simulated real-world complexities such as noise, class imbalance, and high dimensionality.
• To investigate the interpretability of ML models using explainable AI (XAI) tools.

Main Methods:

• Compared five supervised ML classifiers (Decision Tree, K-NN, Random Forest, Linear Regression, SVM) against three QML classifiers (Quantum SVM, Quantum K-NN, Variational Quantum Classifier).
• Utilized five diverse datasets (Iris, Wine Quality, Breast Cancer, UCI HAR, Pima Diabetes).
• Introduced class imbalance (SMOTE, ADASYN), feature noise (Gaussian), and dimensionality reduction (ANOVA); employed SHAP and LIME for interpretability.

Main Results:

• Logistic Regression demonstrated consistent performance across various complexities.
• Quantum Support Vector Machines exhibited notable resilience to feature noise and class imbalance.
• Explainable AI tools provided insights into model decision-making processes.

Conclusions:

• QML models, particularly Quantum SVM, show potential for handling complex, real-world data challenges.
• The study highlights current QML capabilities and limitations, informing the development of generalisable and interpretable ML systems.
• Findings are vital for deploying robust AI in complex, practical environments.