Skip to content

Multiple linear regression models to predict non-airline revenues at DEN airport in R.

Notifications You must be signed in to change notification settings

Drev917/DEN-MLR-Model

Repository files navigation

DEN-MLR-Model

Problem Statement:

Assess patterns and correlation in non-airline revenue factors collected at DEN (Denver International Airport). Using data accrued from January 2012 - June 2017, create a method for forecasting non-airline revenues. Use it to produce forecasts for the months of March through June of 2017, and compare with the real observations to check your model performance.

ScreenShot

First we can look at our goals:

*On top of predicting total non-airline revenue as a comprehensive response variable, we looked at each of the dependent variables independently.

ScreenShot

We then looked at the data as a whole to gain an idea of the different variables and observation values:

ScreenShot

We used the following libraries to perform the EDA:

  • dplyr
  • data.table
  • tidyr
  • janitor
  • reshape2
  • stats
  • ggplot2

ScreenShot

We also looked at seasonality between non-airline revenue and months/years as a timeseries

ScreenShot

For our comprehensive non-airline revenue response variable: TotalRev we chose the following predictor variables as statistically significant and influential towards our regression model:

  • Month
  • Destination
  • Cannabis?
  • UMCSENTLag3

Adjusted Coefficient of Determination: 85.6%

ScreenShot

We noticed non-normality of errors when plotting the residuals and checking for multiple linear regression violations of assumptions.

By performing a log transformation on the response variable TotalRev, we were able to spread our residual plot and correct any violations.

ScreenShot

Lastly, an analysis was run on each of the factors that comprised the response variable in turn looking at residual standard error and ABS of the difference between training and test sets.

ScreenShot

Thanks for reading

Releases

No releases published

Packages

No packages published