Maternal mental health (MMH) during pregnancy is becoming an increasingly important aspect of perinatal care. Depression, anxiety, and posttraumatic stress disorder (PTSD) are significant contributors to maternal and child morbidity but are often missed in standard screening. The goal of this capstone project was to use statistical and machine learning methodologies to identify predictors of mental health risk as defined by these three validated instruments: the City Birth Trauma Scale (CBTS); the Edinburgh Postnatal Depression Scale (EPDS); and the Hospital Anxiety and Depression Scale (HADS).

The study uses data from Swiss pregnant people (n=410) and explores the relationships between sociodemographic factors (age, education, gestational weeks, and marital status) and mental health outcomes among those pregnant people. Both standard regression models and ensemble machine learning models contributed to early identification approaches and revealed potential vulnerability patterns.

The project advances this goal through the development and evaluation of predictive models for each outcome. The findings provide insight into the role of sociodemographic health in psychological well-being.

Bio

Mikel Houston is a Master of Science in Data Science student at Meharry Medical College with a background in biology and a research focus at the intersection of healthcare, cybersecurity, and emerging quantum technologies. Her work centers on developing secure, data-driven systems using machine learning, big data frameworks, and cloud computing. She is particularly interested in quantum-resilient security for healthcare IoT environments, where sensitive patient data must be protected against evolving computational threats. Mikel aims to contribute to advancements in secure healthcare infrastructure through interdisciplinary research combining data science, artificial intelligence, and post-quantum cryptography.

Abstract

Mikel Houston’s capstone explores quantum-resilient security frameworks for healthcare IoT systems by combining Quantum Key Distribution (QKD) with Post-Quantum Cryptography (PQC). The project simulates secure communication between AI-enabled wearable devices, edge gateways, and cloud infrastructures, evaluating performance across key metrics including latency, scalability, and resilience against both classical and quantum attacks. A comparative analysis of classical encryption, PQC-only, and hybrid QKD-PQC approaches highlights trade-offs between efficiency and security. This research informs the design of next-generation cybersecurity architectures, ensuring robust protection of sensitive healthcare data in increasingly connected and intelligent medical environments.

Advisor: Uttam Ghosh, Ph.D.

Abstract

Sarcoidosis is a multisystem inflammatory disease characterized by granuloma formation and variable clinical presentations. Accurate diagnosis is often difficult because its symptoms overlap with other inflammatory and pulmonary conditions, and the absence of reliable non-invasive biomarkers. This study aimed to improve molecular classification of pulmonary sarcoidosis using machine learning applied to integrated multi-cohort gene expression data. Previous sarcoidosis studies relied on single-cohort datasets with limited generalizability. Three publicly available Gene Expression Omnibus datasets (GSE18781, GSE42834, and GSE83456) were combined and harmonized, with a final dataset of 398 samples: 229 healthy controls and 169 sarcoidosis cases, with 16,622 shared genes retained for modeling. The analytical workflow included data preprocessing, batch correction, low-variance gene filtering, statistical feature selection, feature importance analysis, scaling, principal component analysis (PCA), and classification modeling. Six algorithms were evaluated—Logistic Regression, Random Forest, Support Vector Machine (SVM), K-Nearest Neighbors, Naive Bayes, and Gradient Boosting—using accuracy and ROC-AUC as primary metrics, and precision, recall, and F1-score for detailed assessment of top models. Random Forest achieved the highest performance (accuracy 0.76; ROC-AUC 0.85), followed closely by SVM. PCA revealed partial separation between sarcoidosis and control samples, supporting moderate discriminatory power. Feature importance analysis highlighted genes involved in immune and inflammatory pathways. These findings demonstrate that integrating multi-cohort gene expression data enhances the robustness and generalizability of machine learning models for sarcoidosis classification.

Advisor: Naw Safrin Sattar, Ph.D.

Increasing sepsis mortality rate remains a critical global health concern, particularly in intensive care unit (ICU) patients. Therefore, early, and accurate prediction of sepsis outcomes is crucial for guiding timely clinical interventions and reducing preventable deaths. This study aimed to investigate a comprehensive machine learning (ML) approach to accurately predict mortality risk outcomes using clinical characteristics such as laboratory results and underlying comorbidities data from the Medical Information Mart for Intensive Care-IV (MIMIC-IV) collected from the first 48 hours of ICU admission. To predict outcomes the study utilized several ML methods, Logistic Regression, Random Forest and XGBoost. We analyzed data from 16,994 septic patients who were admitted into the ICU. Model performance was assessed using a combination of cross-validation and test set metrics, including ROC AUC, average precision, and accuracy. TheXGBoost model demonstrated the best overall performance, achieving a test ROC AUC of 0.86, average precision of 0.76, and accuracy of 80%. Logistic Regression and Decision Tree models showed moderate predictive capability with lower AUC and precision scores. The results confirmed that ensemble methods, particularly XGBoost, offer superior performance in predicting ICU mortality among sepsis patients in contrast to commonly used traditional scoring tools like SOFA. These findings highlight the potential of machine learning to aid in early risk stratification and informed clinical decision-making in the management of sepsis.

Children’s mental health is increasingly recognized as vital to overall health, yet many external factors outside biological predispositions influence emotional, psychological, and social outcomes. Using nationally representative data from the National Survey of Children’s Health (NSCH), this study explores the extent to which the Socioeconomic and Environmental factors contribute to disparities in mental health outcomes such as anxiety, depression, ADHD, and autism. Logistic regression and Random Forest were applied to assess and predict risk factors. Findings indicate that disparities in income, neighborhood safety, healthcare access, and parental education significantly impact mental health. This research provides evidence-based recommendations to promote mental health equity among children.