Machine Learning in Aviation
DataHub Consulting, Experts in Analytics, Business Intelligence, and Compliance 310 310Read it in 12 minutes
window.dataLayer = window.dataLayer || []; function gtag(){dataLayer.push(arguments);} gtag('js', new Date()); gtag('config', 'G-YFZ1F7T6M6');
window.dataLayer = window.dataLayer || []; function gtag(){dataLayer.push(arguments);} gtag('js', new Date()); gtag('config', 'G-YFZ1F7T6M6');Read it in 12 minutes
As Datahub Consulting are experts in machine learning and analytics, coupled with our expert knowledge of the aviation industry and data, I thought it would be good to write a blog on how machine learning is currently changing airlines and how it will affect them in future.
This is a non-technical blog aimed at anybody interested in how machine learning is affecting this industry. If there is interest in a technical blog on this topic then please get in touch and I we can do this.
Within aviation there is a massive growth in the use of data science in particular machine learning and artificial intelligence. For these people that have heard if these terms but don’t understand the meaning, machine learning is a subset of artificial intelligence. As we go through this blog you will understand more about machine learning as well as the implementation within aviation.
In aviation there is a huge amount of data collected from the aircraft, airline head offices, to GPS data that use ADS-B, and finally phone apps can track a flight in real-time. Phone apps and ADS-B devices can plot an aircraft’s course, altitude, ground speed, outside temperature etc.
Airline use data to make decisions and to help make the business more efficient. Whilst an aircraft is on the ground it’s not making money for the airline. So, the more time that an aircraft is in the air the more efficient the airline is. So how do we improve efficiencies?
Also, in the aviation industry good customer service isn’t enough anymore. Airlines are looking at ways to put their customers, or commonly referred to as “guests” at the centre of the business. Airlines are looking to utilise machine learning techniques that retail have been doing for a number of years. These techniques personalise the customer experience and makes the customer journey as streamlined as possible. So, airlines are looking at their processes through the eyes of the customer. From planning a vacation or flight, through the reservation process, to the airport experience, as well as the flight itself.
To describe machine learning in a non-technical way I would say it’s a method to model historic data to predict outcomes in the future. The machine learning model inputs historic data points that are relative to the problem. These data points are known as “features”. Using the features, the machine learning model creates an algorithm that will predict the outcome and the probability level as a percentage.
With all this in mind, how can data science and machine learning help? Firstly, with any machine learning project you need a problem to solve.
All airlines are looking to save money but also to maximise the service to the customers. There are different ways to save money and one is to reduce the fuel used on a flight. Pilots can do this to an extent with their speed and altitude, but weather also plays a big part in this. For example, if a plane is flying into a headwind then it will need more power (thrust) to maintain a particular speed than if it had a tail wind. So how can machine learning look at optimising the fuel consumption. Qantas are the first airline to have flown a scheduled flight between Australia and London (LHR), a 17-hour trip nonstop. It’s also not surprising that Qantas are also using machine learning within their flight operations.
Let’s look at an example of how using weather could help reduce fuel consumption, also reducing carbon emissions. Within the flight operations team at an airline they plan the flight details of a scheduled flight. This could include the time of the flight, the aircraft used, and also routing etc. Now if they included weather data and machine learning into the decision making, they can reduce fuel consumption. For example, if a machine learning algorithm has weather data that included wind direction, wind speed, any bad weather like storms, then this could play a considerable role. So, if the algorithm recommended that a plane fly’s 5 degrees east of its planned route and would then get the benefit of a tail wind this could reduce the fuel consumption.
Airlines always want to keep the aircraft flying and not to be grounded for any duration in time. Least of all unexpectantly. So, machine learning can support preventative maintenance in this way. Using data from aircraft along with other external datasets a machine learning anomaly detection solution can be created that looks at the data points from the aircrafts mechanical, electrical, hydraulic systems etc and identifies components that need to be changed before they fail and ground the aircraft. Anomaly detection works by identifying the normal state of a component seen in the chart below the yellow data points in the green oval. The oval identifies the normal thresholds based on historic data. Anything that is plotted outside of this threshold would act as an alert. These alerts would be categorised. So just outside the threshold would be a low alert, but a data point (such as the blue triangle below) would be considered a high alert.
Within the aviation industry, an aircraft could do 350 long haul flights over a set period in time. And in that same period another identical aircraft could do 1000 short haul flights. This could have an impact on any of the aircraft components. Also, the geographic location may have an impact on the aircraft components as extreme cold or hot conditions could deteriorate the longevity of the component. So, all this data can be used to identify failure points and to plan maintenance so that these failure points are managed.
Visual to demonstrate how anomaly detection works
Within machine learning there are a number of examples available for predicting flight delays. Flight delays lead to a negative impact on the airline. These algorithms predict the likelihood of a flight being delayed taking a number of factors into consideration, including weather, route, outbound and inbound airports. The algorithm is trained on historical flight delay information from the airline, FAA, historical and forecasted weather, and the current state of the National Airspace System.
Airlines, like hotels, want to provide not only a service but also a personalised experience. Travel isn’t just getting from one country to another, it’s making the little details matter that makes the customer want to travel with the airline.
To provide a personalised customer experience, organisations are putting the customer at the centre of their business. So, from an airline perspective how can they use machine learning to achieve this?
Firstly, by collecting data on customers listed below the airline can get a better understanding of their customers, their needs, preferences, and how the airline can personalise the customers experience.
Examples of datasets required to personalise customer experience:
Once that a profile (or a picture) has been created of the customer then the machine learning model can then make recommendations to the customer. Amazon use this type of machine learning a lot.
This type of machine learning is called a Recommendation Engine.
Targeted Marketing Let’s look at some reported stats:
Based on information from Capgemini
The above information isn’t based on the aviation industry but marketing trends in general, but these could easily be applied to the aviation industry.
Majority of an airline revenue isn’t from the vast majority of the seats (economy), but from the fewer seats used by:
So, with that in mind it’s important to any airline to have a good FFP to increase revenue. Also, airlines need to think about not losing members from the FFP to competing airlines. This is where big data, membership data, and reservation data coupled with predictive analytics can help.
This particular machine learning is known as churn, being able to identify members of a loyalty program, where there is a possibility of them leaving or reducing the services that they require. Examples of data used in this could be:
All of the above datasets can create a picture of the customer. From this picture a prediction could be made of the likelihood of the customer churning. From the prediction engine a score will be generated based on the likelihood of churning and anyone above the score threshold would be considered as medium or high risk of going elsewhere for the services. Using this score the data can easily be visualised in a dashboard.
Machine learning can also give an indication of the time frame that they are likely to churn. So, you have an indication of how quick you need to react.
With machine learning, airlines can look to optimise the fares to keep them competitive to other airline operators. With more airlines available, budget airlines, and the internet, fares can easily be compared by everyone. Historically, to book a flight you needed to visit a travel agent who would search the system and provide options. Now with the internet and all airlines providing direct online sales, competing airlines need to be value for money.
So how can machine learning help with this?
Firstly lets look at it from the airlines prospective, they need to think about selling tickets as high as possible (they are a business with shareholders), and also sell as many tickets as possible. They don’t want to fly with only 50% of the seats taken. They do this by considering:
Based on the above tickets can fluctuate accordingly for maximum sales revenue.
Now let’s look at it from the customers prospective. The customer wants to get a low fare and to achieve this there are two considerations:
The machine learning models will look at features such as:
So, what do I mean about customer preferences? Customers may prefer particular departing airports, overall flight duration. Also, is the flight direct or do I need to get a connecting flight, customer service and facilities. Finally, have I had a bad experience with a particular airline?
All these are customer preferences that need to be considered.
Below illustrates consideration for fare optimisation from both the airline and customer prospective. These considerations would be used in the machine learning.
Please remember, where you use personal information as part of any analytics / machine learning project then the relevant data compliance legislation needs to be taken into consideration.
I’ve been asked a lot of questions about this and for that reason I’ve created a separate GDPR blog. The short answer to this is no. But under the legislation, It will change the way that you use data in machine learning. It’s not aimed at stopping you from creating machine learning models.
So long as you have considered:
The six principles of GDPR Principle
The Rights of the data Subject
And finally, under article 6: Lawfulness of Processing – you have the right to process any personal information within a machine learning model so long as you are doing so lawfully. There are a number of lawful reasons to store and process personal information but for example, have you gained the consent of the data subject, or do you have a legitimate Interest to do so?
For legitimate interest there is a balance test to see if this is a lawful reason for processing the data with machine learning. On the ICO website there is information and a balance test that is very good to use.
If you need help with compliance and GDPR please get in touch as Datahub Consulting have staff that are GDPR Practitioners and can support any project on compliance.
With all this in mind Datahub Consulting are experts in the field of data science and machine learning. We have opened our Aviation Center of Excellence (ACE) where we provide airlines, airports, airport services with end to end project solutions. This includes the information gathering and design, development, testing, deployment and documentation of bespoke solutions. We have experts with aviation knowledge and experience in aviation datasets and data applications. The team comprises of technical consultants, project managers, client support team that can provide data science, machine learning, analytics, dashboards, data compliance and governance solutions.
If you want to talk to us in more depth about our services or the aviation ACE then email us on info@datahubconsulting.co.uk, or call our UK office on 0116 223 0689.
We do not employ salespeople; our team are all experienced technical specialists that can talk you through any of our services.
Contact us