International Conference on Trends in Quantum Computing and Emerging Business Technologies (TQCEBT)
Quantum Enhanced Ensemble Learning for Autism Spectrum Disorder Prediction with Boltzmann Machines and Generative Models
Vishal Peter, P. Renukadevi, S. Annamalai, P. Manikandan
Abstract
This paper suggests a proposal of a quantum-enhanced ensemble learning model to predict Autism Spectrum Disorder (ASD). The model is a hybrid of a Quantum Boltzmann Machine (QBM) branch (learns discriminative features), and a generative model branch (learns distribution), and their predictions are fused together on an adaptive basis (depending on confidence and computational cost). Three ASD screening datasets of public data, Adult (704 instances), Children (292 instances), and Adolescent (104 instances), are used to evaluate the proposed approach. Accuracy is constant at over 95% on all datasets, with accuracy is 96.5% (Adult), 97.1% (Children), and 95.8% (Adolescent). The accuracy measures between 95.2% to 96.4%, and the recall between 96.4% to 97.8% portrays a high predictive accuracy. The F1-Score is similarly high, with the range of 95.8% to 97.2% and AUC of 97.5% to 98.7%, showing the model has a strong capacity in distinguishing between cases of ASD and non-ASD. In addition, adaptive fusion method enhances the robustness of the prediction by dynamically adjusting the input of the separate branch, which has raised the general model reliability. The results show that quantuminspired techniques are useful in better ASD classification and they are computationally efficient (inference times of 120ms-130ms). Among other weaknesses are increased complexity when working with larger datasets, the need to further develop explainability and fairness, which are noted, and future research is described.