Analysis of FLTWinds Data using a Neural Network Based Approach

1
Analysis of FLTWinds Data using a Neural Network
Based Approach Haimonti Dutta CIS Department
,Temple University
2
FLTWinds - The Flight and Weather Information
and Decision Support System

• Features
• Aviation weather data management
• Creation of advanced aviation weather products
• Weather management and alerting services
• Flight tracking and display services
• Flight following and alerting services
• Sophisticated mapping and display tools
• User interface that combines both flight and
  weather information on a common graphical display
  
•  

3
Collection of the Data
A View of the Database Schema
Attribute name Attribute Definition
A_IN Actual Gate in time
A_IN_SRC Source of time stored in A_IN column
A_OFF Actual wheels off time
A_ON Actual wheels on time
A_ON_SRC Source of time stored in A_ON column

• The tables used in the Database schema are
• Flight
• Plan
• Plan_Point
• Tracking
• Airline
• About 3 GB of data is collected per month.

4
Steps in Data Preprocessing

• Attributes required to build the database
• Removal of uninteresting attributes like
  Route_Date, Plan-Number, Plan_Time etc
• Removal of attributes for which data was not
  available. For e.g SUA_ALERT, WX_ALERT,
• FUEL_REMAINING.
• Some of the major attributes chosen include
• FLEET_ID, DIVERT_TIME, PLAN_DISTANCE, ALERTS,
  ARRIVAL_DELTA,
• DEPARTURE_DELTA, HOLD_TIME, MAX_OFF_RTE,
  DISTANCE_DELTA etc.
• In all, 26 attributes were chosen for the final
  data processing.
• Chosing airport hubs for data analysis(A data
  reduction step)
• After the attributes were chosen, the next step
  was to choose the 5 major airport hubs in USA
• Including the Boston Logan Intnl. Airport(BOS),
  Baltimore Washington Intnl Airport(BWI),
• Chicago OHara Intnl Airport(ORD),
  Dallas-Fortworth Intnl airport(DFW), Denver Intnl
  airport
• (DEN).(Based on ranking of busy airports-
  http//airtravel.about.com/library/news/airports/b
  larptnewsRankings.htm)
• Data was collected for all aeroplanes which were
  coming
• flying into these hubs on the
• Specified dates.

5
Data Sets
Number of records analysed -
Airport Id Number of records
BOS 1447
BWI 545
DEN 7306
DFW 1552
ORD 4791

• For each of these airport hubs, a neural network
  classifier was built for identification of two
  classes.
• Flights on-time
• Flights not on-time (early/late).

6
Distributions of Arrival times of Flights at the
airport hubs chosen
BOSTON
CHICAGO
DENVER
BALTIMORE- WASHINGTON
7
Results
Classifier Accuracy
Airport Accuracy
BOS 99.53
ORD 88.12
DEN 92.57
BWI 69.0
DFW 56.57
Plot of the Accuracy
8
Experiments to be done
According to domain experts, the displacement
from the actual route of a flight is an
important Attribute that needs to be analyzed.
Initial examination reveals the following
patterns for the max_of_rte attribute.
Distribution for max_of_rte
BOSTON
CHICAGO
DENVER
BALTIMORE - WASHINGTON
9
Future Work

• Examination of other attributes including
  departure_delta, alerts, diversion_alert,
  hold_alert
• Examination of the performance of air bus and
  Boeing aircraft
• Development of a linear regression model for
  estimation of arrival times of aricrafts
• Patterns in delay of flights.

• References
• Neural Networks, A comprehensive foundation by
  Simon Haykin
• FltWinds software in use at Lockheed Martin
  corporation
• Domain experts including Dr. WolfGang, Dr. Biju
  Kalathil, Dr. John Carlsen, Rusty Bell.

10
Questions ????