Implementing Bayesian Inference to analytical problems.
Using tweets, we predict whether Trump or Clinton wrote them. NLP was used as to prepare unstructured data for modeling. We compare how a Naive Bayes classifier performs using Frequentist Statistics and Bayesian Statistics (Laplace Smoothing). We also employ a new algorithm for text analytics: TF-IDF.
Given a Gamma prior and exponentially distributed data points, we derive the marginal and predictive distribution of the data. We also propose a mixture framework for combining prior beliefs.
A Generalized Linear Model, concretely, a logistic regression, is estimated to predict whether a patient had a heart disease or not. Feature selection is carried out with L1 Regularization or LASSO regression and then Frequentist and Bayesian Inference are compared.