1. PREDICTING Flight Delays
Washington DC area airports
BIT Applied Business Intelligence and Analytics - Spring 2017
Group 3: Alexandra Robleto, Caitlin Fernandez, Lucas Cameron, Kevin Sherman

2. Project Summary Establish business need Collect data
Understand flight data
Prepare data for modeling
Create predictive models
Measure models ability to predict delays
Evaluate findings
Make recommendations

3. data Data source Data content Data understanding Data preparation
Data content
Calendar Year 2016
All Flights from Washington DC area airports: BWI, IAD, and DCA
Data understanding
Available variables
Relationships
Data preparation
Missing values
Outliers
Redundant variables

4. Preliminary findings – Average delays
Washington Reagan airport had more delays on average
Certain airlines experience more delays and the airlines are different depending on the airport
Highest delays during summer months June and July, followed by December

5. Preliminary findings – Average delays by time of day
Flights departing before noon tend to arrive early
Delays tend to get worse after noon

6. Preliminary findings – total delays by reason
Late aircraft was the most common cause of flight delays in 2016, followed by carrier delays
Security issues were least likely to cause delays, and weather was not an important cause of delays either

7. Preliminary findings – canceled flights
While weather was not an important cause of delay, it did contribute to most flight cancelations, especially in January 2016.
While Southwest Airlines cancelled the most flights, they also had the highest number of flights.
On the other hand, Delta Airlines had fewer cancelled flights compared to their number of flights.

8. Preliminary findings – diverted flights
Summer months had highest delays, and also the most diverted flights, regardless of the airport.

9. Predictive modeling process
Training and Validation
Logistic Regression
Classification Tree
Neural Network

10. Evaluation of predictive models
Receiver Operating Characteristic (ROC) curve
The closer it gets to the top left corner the better
Area Under the Curve (AUC)
The closer to one the better
LR: Logistic Regression model
DT: Decision (or classification) Tree model
Neural: Neural Network model

11. Evaluation of predictive models
Fit or Accuracy
Rsquare: the higher (closer to one) the better
Misclassification Rate: the lower the better
Lift curves
Model performance as opposed to guessing
The higher the better
LR: Logistic Regression model
DT: Decision (or classification) Tree model
Neural: Neural Network model

12. Conclusion and recommendations
Best model based on evaluation techniques
Classification Tree
How the model and insights address the business need
Possible delays identified based on flight booking information
Alternative flights presented
Ways to improve the model
Include more inputs
Increase amount of data