Quantum Support Vector Machines
Particularly when working with complicated and frequently unbalanced healthcare datasets, recent demonstrate the substantial potential of Quantum Machine Learning (QML) techniques, in particular Quantum Support Vector Machines (QSVMs), to increase the precision and dependability of disease detection. Using datasets for diseases including diabetes, heart failure, and prostate cancer, quantum models are compared to conventional machine learning techniques to show a definite advantage for quantum approaches in some crucial areas.
In contemporary medicine, identifying diseases accurately and early is still a major difficulty. One major barrier is the occurrence of imbalanced datasets in the medical field, where there are sometimes many more cases of an illness (positive cases) than healthy cases (negative cases). When confronted with this imbalance, traditional machine learning methods often perform worse. Scientists are working to find out if quantum computing, which makes use of concepts like entanglement and superposition, may improve pattern recognition in these difficult situations.
Quantum Neural Networks (QNNs) and Quantum Support Vector Machines (QSVMs) were compared to well-known classical algorithms like Logistic Regression, Decision Trees, Random Forests, and classical Support Vector Machines (SVMs) in a comparative analysis described in research by Tudisco et al. and published by Quantum Zeitgeist. Prostate cancer, heart failure, and diabetic healthcare datasets were used to evaluate how well quantum techniques might be able to get over the drawbacks of unbalanced data.
Across all evaluated datasets, the QSVMs continuously beat QNNs and classical models. This implies that quantum models have a significant edge in challenging classification tasks. This superiority was found to be especially noticeable when datasets showed notable imbalance, which is a typical feature of many healthcare issues. With extremely unbalanced datasets, like the Heart Failure dataset, where traditional algorithms frequently fail to obtain high recall, quantum models showed a distinct benefit. Quantum models, particularly QSVMs, demonstrated higher performance in properly recognising positive instances (high recall) in these contexts, suggesting the possibility of increased diagnostic accuracy in difficult clinical scenarios. The more class disparity there is, the more advantageous using quantum models seems to be.
Although QNNs displayed encouraging precision scores, their propensity to overfit the training data limited their usefulness. Reduced generalisation performance on unseen data results from overfitting, which happens when a model learns the training data too well and captures noise and particular features instead of fundamental patterns. QSVMs, on the other hand, demonstrated higher resilience and reliability. The capacity of QSVM to reliably identify positive cases in a variety of clinical contexts is demonstrated by its consistently high recall across all datasets. Future research on QNNs should concentrate on reducing overfitting, perhaps by investigating different circuit layouts and improving hyperparameter optimisation.
In order to improve diagnostic performance over traditional SVM, a particular “Quantum Support Vector Machine for Prostate Cancer Detection: A Performance Analysis” looked directly into the use of QSVM for prostate cancer diagnosis. Prostate cancer treatment effectiveness and improved patient outcomes depend on early detection. Although the interpretation of biomedical data has improved with classical SVMs, these models are limited by the big, high-dimensional datasets that are common in this domain. Using quantum concepts like superposition and entanglement, QSVMs are emerging as a cutting-edge technology that can manage multidimensional data and possibly speed up operations.
The Kaggle Prostate Cancer Dataset, which at first had 100 observations with 9 variables, including clinical data and diagnostic results, was used as part of the approach for the prostate cancer. During preprocessing, the RandomOverSampler technique was used to rectify the initial class imbalance in the dataset. In order to improve feature comparability and get the data ready for quantum encoding, it was also normalised using MinMaxScaler and standardised using StandardScaler. Training (80%) and testing (20%) subsets were created from the processed data, which had been oversampled to 124 samples.
The adoption of a quantum feature map architecture, specifically the ZZFeatureMap with full entanglement, which was carefully chosen and assessed to match the particulars of the dataset, was a crucial component of the QSVM technique. By encoding conventional data into quantum states, this feature map enables the quantum system to use entanglement to describe intricate data correlations in high-dimensional regions. In QSVM, the inner product (overlap) between the quantum states that represent data points is estimated in order to construct the kernel function, which is essential for SVM classification. A quantum circuit is used for this estimation, and the likelihood of seeing the starting state is measured.
The prostate cancer experimental findings offer convincing insights:
- Different patterns were found by kernel matrix analysis. A highly connected feature space was suggested by the high similarity values between several data points displayed by the RBF kernel of the classical SVM. On the other hand, the QSVM’s ZZFeatureMap produced a feature space that was more dispersed and had fewer high off-diagonal values. This suggests that the special characteristics of the quantum feature space allowed for improved class distinguishability.
- QSVM greatly outperformed traditional SVM (87.89% accuracy, 85.42% sensitivity) on the training dataset, achieving perfect scores all around (100%). This demonstrates how well the quantum feature map separates classes without overlap during the training stage.
- Both models had an accuracy of 92% on the test dataset.
- On the other hand, QSVM outperformed SVM in important medical diagnostic parameters, achieving 100% sensitivity and 93.33% F1-Score on the test data, while SVM’s sensitivity and F1-Score were 92.86% and 92.86%, respectively.
- Crucially, the test data showed no False Negatives (missed malignant instances) for the QSVM model. In contrast, there was only one False Negative in the SVM model. In medical situations, where a false negative might have serious consequences and result in an illness that goes misdiagnosed and untreated, QSVM’s high sensitivity is crucial. This is made possible by the quantum feature mapping, which improves class separation and permits more intricate representations.
- Despite its excellent performance on the test set, the QSVM model may have overfitted to the training data, according to cross-validation research, which showed that the SVM model was somewhat more stable across various data subsets than the QSVM.
The discussion highlights how QSVM’s improved sensitivity and F1-Score are especially useful in medical diagnostics. This improved performance is a result of the quantum feature mapping’s capacity to produce a distinct, dispersed feature space, particularly when separating complex data points. The study is thought to be the first to classify datasets related to prostate cancer using the QSVM algorithm.
