Project

Approaches

Dataset – The U.S. Department of Transportation (2015 Flight Delays and Cancellations).
Almost 6 million rows and 31 columns.
The data of January will be used.

Data Cleaning
- Airline
- Original airport and Destination airport
- Year, Month, Day, Day of week
- Scheduled departure and Scheduled arrival
- Departure time and Arrival time
- Departure and Arrival delay

Filling Factor
- Examine how complete the dataset is.
- Variables filling factor is quite good (> 97%).
- Remove the entries that contain missing values.

1. One Airline, One Airport

Pitfalls
- Insufficient Statistics
- Extreme Delays
Polynomial degree
- Splitting the dataset to test and training.
- Bias introduced by the separation of the dataset.
- Splitting the dataset to test, training and validation.
- Selection by cross-validation.

2. One Airline, All Airports

Prediction
- Data of the first 3 weeks of January.
- Run model on the training set.
- Predict the fourth week of January
- Compute the MSE score between predict and actual delay.

Rank of Airlines
- Calculate factors affect the rank of airlines
  1. Highest ratio of (Operated flights)/(Scheduled flights)
  2. Flight speed
  3. Average arrival delay
  4. Flight volume
  5. Taxi in and out Time
- •Our rank method
  - Score = a/(1+b)
    - a = (RATIO_OP_SCH) * (FLIGHT_SPEED) * (FLIGHTS_VOLUME)
    - b = (ARRIVAL_DELAY) * (TAXI_IN) * (TAXI_OUT)

Where,

RATIO_OP_SCH is (Airline operate) / (Scheduled departure);
TAXI_OUT time - The time duration elapsed between departure from the origin airport gate and wheels off.
TAXI_IN time - The time duration elapsed between wheels-on and gate arrival at the destination airport.

Fig1. Ratio of operated and scheduled flights for each airline in increasing order (Top-left)

Fig2. Flight speed (Top-right)

Fig3. Mean flight delays (Arrival : Green, Departure : Yellow) (Bottom-left)

Airlines Recommendation
- Users would like to choose one of the three or four candidates in filter list.
- Airlines recommendation is based on rank of airlines.
- Use rank of airlines to recommend the airline for users.

Predict Delay Results on Model 1

Predict Delay Results on Model 2

Linear Regression

Polynomial Regression

Test result in Fig. 5 and 6

Rank and Recommendation of Airlines

Airlines Recommendation

Data analysis algorithms are applied to predict flight delay.
Airlines are ranked for recommendation purpose.
In model 1, cross-validation can avoid bias introduced by splitting data.
In model 2, compared with linear regression, polynomial regression with ridge regression is the wining method with MSE (54.99).
Include almost all the factors to rank airline for users.

CODE

WEB PAGE

https://sites.google.com/view/ee608

Google Sites

Report abuse