Predicting NBA Players Salaries with Bayesian Linear Regression

Coding details and project report can be accessed here

Project Goals

• Get hands on experience with Markov Chain Monte Carlo methods
• Understand the difference between Bayesian Approach vs. Frequentist Approach
• Test on No U-Turns Sampler (NUTS) method (Hamiltonian Monte Carlo method)

Solutions

• Built two bayesian linear models one with Metropolis-Hastings and the other with No U-Turns Sampler(NUTS) sampling method.

Results

• Selected features

• Evaluation scores

• The trace plot and the posterior distributions of parametersof the model with 1000 samples for illustration purpose

Abstract

As one of the most famous and useful statistical models, linear regression has been studied by researchers from every possible angle. Among the various perspectives that people look at the linear regression problem, two genres stand out and have been "competed" for decades, which are Bayesian and Frequentist. In this project, I aim to build a linear regression model to predict NBA players' salaries from both the Bayesian approach and the Frequentist approach. To be more specific, I want to study the difference between the two approaches in terms of the result and the interpretations of the parameters. In the next section, I am going to give a brief introduction to the data I collected for this project and the data preprocessing I have done.