Statistical Methodology Commons^™

A Comparative Evaluation Of Data Imbalance Handling Techniques In Machine Learning Models For One-Year Mortality Prediction In Liver Cirrhosis, Sumiya Hasan Trisha

UNF Graduate Theses and Dissertations

Liver cirrhosis is associated with substantial morbidity and mortality, making one-year mortality prediction a clinically relevant problem. Using a liver cirrhosis dataset as the motivating application, this thesis evaluates five machine learning classifiers—Logistic Regression, Random Forest, XGBoost, LightGBM, and CatBoost—under five class-imbalance handling strategies: Baseline learning, Random Oversampling, SMOTE-NC, ADASYN, and Cost-Sensitive Learning. Hyperparameter tuning was conducted using randomized search, and predictive performance was assessed over 200 iterations of Monte Carlo Cross-Validation using Accuracy, Precision, Recall, Fl-score, and ROC-AUC.

The results suggest that imbalance-handling strategies can materially affect predictive performance, particularly recall. Because the outcome of interest is death within …

Variance Shrinkage In Dunnett-Type Multiple Comparisons With Missing Data, Md Habibullah

UNF Graduate Theses and Dissertations

Dunnett’s procedure is widely used for comparing multiple treatments with a control, but its application becomes challenging in the presence of missing data and multiple com- parisons. An improved Dunnett-type procedure addresses this by using multiple imputation under Rubin’s framework and constructing unified confidence intervals based on a multi- variate t distribution, allowing valid simultaneous inference while controlling the family-wise error rate (FWER). This work further extends the method by incorporating shrinkage-based variance estimation. Specifically, individual group variances are shrunk toward a common value to improve stability. This approach is particularly effective when group variances are similar or moderately different, …

Ridit-Based Adaptive Allocation In Two-Stage Clinical Trials With Binary Outcomes, Dewan Fahim

UNF Graduate Theses and Dissertations

This study explores better ways to assign patients to treatments in clinical trials with binary outcomes, such as success or failure. Adaptive methods are used to learn from early results and adjust treatment assignments during the trial, helping more patients receive better performing treatments while maintaining reliable conclusions. We focus on trials comparing multiple treatments using a two-stage design. In the first stage, several treatments are tested to identify the most promising one; in the second stage, that treatment is compared with a control. Unlike traditional equal assignment, we use adaptive allocation in the second stage to make better use …

A Comparative Study Of Classification Methods For Healthcare Analytics, Xueting Zhao

UNF Graduate Theses and Dissertations

This thesis presents a comparative study of logistic regression, Linear Discriminant Analy- sis (LDA), and Quadratic Discriminant Analysis (QDA) for binary classification in healthcare analytics, integrating theoretical derivation, simulation, and real-data application. A facto- rial simulation study crosses the covariance structure (equal vs. unequal), predictor correla- tion (ρ ∈ {0, 0.5, 0.9}), dimensionality (p ∈ {2, 5, 10}) and sample size (n ∈ {50, 100, 200}) across 54 scenarios with 1,000 Monte Carlo replicates each. Three main findings emerge. Logistic regression and LDA are nearly interchangeable when the assumption of equal-covariance holds. QDA achieves substantially better discrimi- nation when class-specific …

Two-Stage Response-Adaptive Randomization Designs For Multi-Arm Trials With Normal Outcome, Tanjin Tamanna Happy

UNF Graduate Theses and Dissertations

This study focuses on improving how clinical trials compare new treatments with a standard treatment, when the response is quantitative (normally distributed). In a common two-stage design, several new treatments are first evaluated, and the best-performing one is selected if it appears better than the standard. In the second stage, this selected treatment is compared again with the standard using additional data to confirm its effectiveness. This approach is known to be efficient in terms of accuracy and sample size savings. We extend this design by introducing an adaptive method for assigning patients to treatments in the second stage. Instead …

Upper-Sided Ewma-Based Distribution-Specific Tolerance Limits, Owen Visser

UNF Graduate Theses and Dissertations

Tolerance limits are constructed from sample data to ascertain if a proportion of a process is within specification limits. There exists multiple methods of calculating the sample size requirements for tolerance limits under various assumptions. In this research, a distribution-specific algorithm that utilizes the exponentially weighted moving average technique (EWMA), first introduced by Sa and Razaila (2004), is reconstructed. The algorithm is used to calculate the required sample sizes for continuous construction of upper-sided tolerance limits. The sample sizes and intervals constructed from them are compared to three existing methods for various distributions. The distribution-specific algorithm was observed to reduce …

A New Right Tailed Test Of The Ratio Of Variances, Elizabeth Rochelle Lesser

UNF Graduate Theses and Dissertations

It is important to be able to compare variances efficiently and accurately regardless of the parent populations. This study proposes a new right tailed test for the ratio of two variances using the Edgeworth’s expansion. To study the Type I error rate and Power performance, simulation was performed on the new test with various combinations of symmetric and skewed distributions. It is found to have more controlled Type I error rates than the existing tests. Additionally, it also has sufficient power. Therefore, the newly derived test provides a good robust alternative to the already existing methods.

Comparing Group Means When Nonresponse Rates Differ, Gabriela M. Stegmann

UNF Graduate Theses and Dissertations

Missing data bias results if adjustments are not made accordingly. This thesis addresses this issue by exploring a scenario where data is missing at random depending on a covariate x. Four methods for comparing groups while adjusting for missingness are explored by conducting simulations: independent samples t-test with predicted mean stratification, independent samples t-test with response propensity stratification, independent samples t-test with response propensity weighting, and an analysis of covariance. Results show that independent samples t-test with response propensity weighting and analysis of covariance can appropriately adjust for bias. ANCOVA is the stronger method when …