Project 1 : Regression models for US primary election prediction

The 2016 US PresIdential Election data was colllected. The dataset includes a selected set of counties in US and their demographic information. It also includes the votes received for Hillary Clinton. Methods : Regression model to predict the percentage of votes of Hilary Clinton in primary election for each county in USA Accuracy metrics : WMSE

Table of contents

1- Simple Regression Model (5 predictors/covariates with simple variable transformation)

-Model fitting

-Results Analysis

-Model Checking and Diagnosis

-Outliers

-Collinearity Diagnosis

2- Full Regression Model (multiple linear regression models with feature engineering, quadratic, interaction effects, or indicator variables)

-Full Run

-Greedy search/Forward selection/Backward selection/Combinations

Project 2 : Forecasting Highway Car Volumes

Table of contents

1- Regression on time

-1.1 Regression on time

-1.2 Diagnostic check

-1.3 Model interpretation

2- Exponential Smoothing

-2.1 Naive application of Exponential smoothing

-2.2 Verification of model choice, upsides and downsides

-2.3 Holt-Winters exponential smoothing

3- Free form forecasting

-3.1 Data cleaning

-3.2 Linear regression

-3.3 Linear regression with ARIMA errors

-3.4 Further Improvements