Big Data in Aviation Industries

Hello friends! I was very busy in last few weeks so couldn’t come up with the new tutorials. Here is blog post I found which is very interesting and will give you brief idea about the aviation industries. What is overbooking. But I think in Airlines uses different overbooking strategy in India because travelling in flight is still luxury and flight cancellation charges are very high. So the number of people that show up for their flight is high compare to other developed countries.

Following blog post can be found here!

Did Big Data Cause United Airlines to Drag a Screaming Passenger Off a Plane?

As a frequent traveller, I have seen some of the logistical challenges involved with running an airline — on a micro-scale. I fly several times per month for work, and I’m a part of the reason that things are so complicated.

 

Inevitably, meetings end early or run late, and so I find myself either trying to get on an earlier flight to get home sooner or going on stand-by for a later flight. Managing passengers that change their plans in unpredictable ways is just one of the many logistical challenges that airlines face.

Unfortunately, the airlines do not always handle their challenges with grace, as demonstrated by this story that I read this morning. A particular flight was overbooked, and nobody wanted to give up their ticket in exchange for a flight credit. Of all the possible options, United Airlines decided to have law enforcement drag at least one unlucky passenger off of that flight. Even more unfortunately, a video of the incident ended up on YouTube.

With the airlines, there’s always plenty of room to find fault; e.g., flying is miserable because the airlines rely on your discomfort to sell premium seats, ticket pricing algorithms are incomprehensible, there are draconian fees applied to everything and inflexible ticket policies make change nigh impossible. The list of [reasonable] complaints is a very long one.

In this particular case, airlines have a very good reason for overbooking flights: it makes good financial sense. Selling only as many seats as there are on the airplane is not only bad for business, but it’s bad for the environment, and bad for passengers too.

How Airlines Should be Using Big Data

Airlines depend on Big Data, collecting mountains of data and information about all their customers and their behaviour. They know when you fly, where you fly, what seats you prefer, where your home is, and so much more. Airlines should (and sometimes do) use data science to maximize your comfort, happiness, and ultimately their own bottom line. Making the customer happier is good for business. Airlines should be using machine learning to solve their problems. Machine learning is just using historical data in order to build a model that can make predictions about future events. For example, since my airline has flown thousands of flights in the past, my data should tell me something about how likely tomorrow’s flight is to be overbooked.

Estimate the number of people that aren’t going to show up for a flight

Maybe the easiest problem for the airlines to solve is estimating the number of people that aren’t going to show up for their flight. The most naive solution to this problem would be to just take the average number of people that haven’t shown up for that flight historically. More complex solutions might make adjustments for seasonality; e.g., winter months probably mean more no-shows due to road conditions. The most sophisticated approaches might include analyzing traffic and weather data to forecast cancellation and no-show rates. Anecdotally, the number of overbooking situations that I’ve seen has certainly decreased over the last few years, so airlines may be getting more sophisticated in their approach here.

Estimate the right offer to make when you are overbooked

Even with the most sophisticated models, there will always be some unpredictability in the system. Fortunately, overbooking is only really a problem if you can’t find someone who is willing to take a later flight. Airlines commonly incentivize people to take a later flight by offering them flight vouchers or even cash to do so.

In this case, though, that didn’t work. The airlines can pay quite a large sum (over $1,000 in cash) to passengers to convince them to take a later flight. I can’t imagine that none of the people on that plane were willing to give up their seat for $1,000. My guess is that the airline just didn’t offer enough.

Machine learning could easily be used to predict the number of people who would be willing to give up their seats for a given compensation rate. That optimization problem then becomes pretty straightforward.

Identify cases where a supervisor is needed

Given the sheer number of incidents that happen at airports, every airline should have an automatic system that alerts a supervisor whenever certain situations arise. If, for instance, a gate agent is having trouble finding passengers to give up their seats on an overbooked flight, a text message should be sent automatically to a supervisor. If passengers have already boarded the plane, but some of them have to be removed, a supervisor should be notified immediately and automatically.

Get rid of change fees altogether

Certainly in the case of the big three airlines, there is more than enough data for them to estimate the number of people who will change their flights. If any one of them would retire this policy — which is really a relic of the past — they would surge ahead of their competitors in a dramatic way. At the very least, why not waive change fees on flights that are already overbooked?

Identify WHY people don’t want to give up their seats

It’s powerful to be able to identify those passengers that are likely to be open to an offer to give their seats up, but it’s even more valuable to understand why your customers do what they do. Data science can give you this as well. Credit agencies have been informing customers why their credit scores went down for years — mostly driven by regulation. That same technology should be used to help airlines understand why their customers make the choices that they make. (As far as I know, DataRobot is the only machine learning platform that makes these reason codes available for every model out-of-the-box)

Update these models all the time

None of the predictive models that airlines build are going to last long because the world is constantly changing. People change, their habits change, locations change, and that means that these models will wear out quicker than many other types of models. Being able to rapidly iterate on existing models to make improvements and updates is key to staying ahead.

So why don’t they do it?

The fact is that doing data science is hard. It’s hard to find people with the skills to do the math and coding, who also know the airline business well enough to be successful. Not only that, but it’s hard for business people and executives to identify the real opportunities in their business to leverage this technology. If they don’t understand it, they won’t prioritize it, and it won’t get done. I’ve seen it happen at hundreds of companies, and the airlines are no different.

Automation is the solution

Hand coding predictive models is an antiquated approach to modeling in most cases. Tools like DataRobot make it possible for business analysts to safely build predictive models in a fraction of the time that a data scientist could. That means the backlog of projects gets cleared fasted, the artificial intelligence that will avoid incidents like the one discussed above get build, and competitors get left wondering where it all went wrong. Automation is the difference between building a few models per year and implementing a few hundred AI solutions every year. It’s the difference between dragging your customers off of your plane and dragging your giant bags of money to the bank.

by Greg Michaelson, Director of DataRobot Labs

 

Originally published Did Big Data Cause United Airlines to Drag a Screaming Passenger Off a Plane?

 

Please let me know your suggestion, thoughts in the comment! Also share with you friends.

Thanks fo reading and keep visiting Analytics Tuts for more tutorials.

 

 

Leave a Reply

Your email address will not be published. Required fields are marked *