QUANTUM INSPIRED HYPERPARAMETER TUNING OPTIMIZATION FOR COMPUTATIONAL PREDICTION OF MENTAL HEALTH DISORDERS

Abstract

Mental health has a deep effect on human health affecting individuals in physically, emotionally and socially. Poor mental health can lead to numerous physical, emotional and behavioral consequences which can significantly impact overall health and quality of life. So, Mental health disorders (MHD) prediction is essential in the field of research for early identification and intervention, improving treatment efficacy, reducing stigma, improving quality of life, addressing Comorbidities, etc. The Mental Health Disorders (MHD) can be tracked using a range of methods including: Quantum Computing (QC), Federated Learning (FL), Machine Learning (ML), Deep Learning (DL), and Edge Computing. In this study we focus on the use of Quantum ML (QML) to predict MHD as Anxiety (AX), Depression (DP), Loneliness (LL), Stress (ST), or Normal (NR). For this study we will evaluate three models: Quantum Gradient Boosting (QGB); Quantum Support Vector Machine (QSM); and Hybrid Quantum Classical Model (HQM). Each model will be compared based on Accuracy (A in %); Precision (P in %); Recall (R in %); F1 Score(F1S in %); Training time (TT in secs); Memory Usage (M in MB); & Model Size (MS in KB) with 80:20, 70:30, & 60:40 Training Testing Ratios (TTR in \%) before and after applying Grid Search (GS) based Hyperparameter (HP) Tuning (HPT) optimization to identify if there is a statistically significant improvement in performance. Additionally, the receiver operating characteristic (ROC) curve will provide TPR & FPR for each model; while Boxplot & Heatmaps will visually represent these parameters. Analysis of results show that the performance of HQM is higher on average (A, P, R, F1S) than A, P and TT with averages of 94.1, 94.0, 94.1, 94.2 and 4.49 for both HQM before and HQM after HPT testing. However, QSM performs better in terms of M and MS with values 1274.41 and 88.54, and 1269.36 and 80.53 accordingly.