| In this dissertation, we consider some Bayesian and multivariate analysis methods in statistical machine learning as well as some applications of Bayesian methodology with differential equation models to study dynamics during co-infections by Leishmania major and Leishmania amazonensis based on longitudinal data.;First, we developed a new MCMC algorithm to integrate the curvature information of a target distribution to sample the target distribution accurately and efficiently. We then introduced a Bayesian Hierarchical Topographic Clustering method (BHTC) motivated by the well-known self-organizing map (SOM) using stationary isotropic Gaussian processes and principal component approximations. We constructed a computationally tractable MCMC algorithm to sample posterior distributions of the covariance matrices, as well as the posterior distributions of remaining BHTC parameters. To summarize the posterior distributions of BHTC parameters in a coherent fashion for the purpose of data clustering, we adopted a posterior risk framework that accounts for both data partitioning and topographic preservation.;We also proposed a classification method based on the weighted bootstrap and ensemble mechanism to deal with covariate shifts in classifications, the Active Set Selections based Classification (ASSC). This procedure is flexible to be combined with classification methods including support vector machine (SVM), classification trees, and Fisher's discriminant classifier (LDA) etc. to improve their performances.;We adopted Bayesian methodologies to study longitudinal data from co-infections by Leishmania major and Leishmania amazonensis. In the proposed Bayesian analysis, we modeled the immunobiological dynamics and data variations by Lotka-Volterra equations and the linear mixed model, respectively. Using the posterior distributions of differential equation parameters and the concept of asymptotic stable equilibrium of differential equations, we successfully quantified the immune efficiency. |
