Flight search engines have become less reliable
In the week of May 30 to June 3, the average price of a round trip ticket in the United States was $ 408, a hundred dollars more than in the same period in 2019, according to data from the Hopper airline ticket application. The rise in prices is partly due to the huge growth in demand from people who, as the pandemic fades, are now fed up with staying at home. Another reason is the high price of fuel, which has risen due to the war in Ukraine. Then there is the lack of workers in the air travel industry. Add to this a flurry of flight cancellations and schedule changes, caused by weather conditions and a whole host of other factors, that explains the strange moment air travel is going through.
In normal times, people often turn to air fare forecasting services to guide their choices. These tools, developed by companies like Hopper, Kayak, Google Flights, Skyscanner, and more, are based on machine learning algorithms. It is one of the first projects in the Big Data sector. The platforms are educated on the arcane rules of airfares and trained with a large amount of historical data, which they then leverage to tell customers the right time to snag the best price for a ticket. Typically, these tools tell a potential traveler whether the prices for the desired route are high or low, or somewhere in between. The more sophisticated services also add a tip: buy now or wait.
But the unusual and peculiar situation in the air travel industry is also reflecting on price prediction services, say some of the executives of these companies . This means that even buyers who are more familiar with the technology may end up paying a little more than the optimal price for a flight. To passengers, buying a plane ticket may seem like a mix of magic and luck, and the current unpredictability could add a hint of confusion and frustration when planning a trip.
How they are determined (and expected) prices WiredLeaks, how to send us an anonymous report Airlines determine airfares with a mixture of art and science. A host of data analysts hired by companies are tasked with predicting who will want to go where and when, and setting timetables, routes and prices accordingly. Even after an airline has set the price of a route, the passenger seated in seat 18A may have paid hundreds of euros more for their journey than the person seated in seat 18B. This is often due to a system called bucket making (literally, “fare bucket”) in which a group of seats are sold for a certain price. Once sold out, more seats are offered for sale at a different price. Automated systems also play their part: if one airline lowers the prices of a route, another carrier could notice the change and immediately reduce its prices.
This causes the companies' price predictions aerial becomes a bit like a fight between spies, in which one giant system tries to predict what another will do. For those who build price prediction algorithms, however, the results are usually consistent: "When they try to predict prices, it is as if the data scientists are looking at a black box," explains Oleksandr Kolisnykov, content strategist at the software company. Altexsoft, which has developed price prediction tools. In the end, however, it doesn't necessarily matter: "In reality we don't know all the reasoning of airlines and the factors that influence current prices, but we can look at the history and make predictions," he adds.
The effects of problems in the sector However, the troubled years of the pandemic have made everything more complicated. Oren Etzioni is now the CEO of the Allen Institute for Ai, but in the early 2000s he built - and later sold to Microsoft - one of the first air fare prediction tools. The forecasting algorithms are quite precise in recalibrating the weight of the different factors according to the changes occurring in the world and, explains Etzioni, "they have the ability to automatically adapt using fresher data". However, it could take some time: days, if not weeks, according to Etzioni.
See more Choose the sportsgaming.win newsletters you want to receive and subscribe! Weekly news and commentary on conflicts in the digital world, sustainability or gender equality. The best of innovation every day. It's our new newsletters - innovation just a click away.
Arrow Google Flights helps customers find the cheapest tickets for the routes and dates they prefer. Since the spring of 2020, however, the search engine has significantly reduced the number of so-called "predictive insights", the service that predict when prices could rise or fall. Overall, Google Flights is aiming for 90 percent accuracy in its predictions, says Eric Zimmerman, director of travel products at Google. "As the volatility of air fares increases, it has become more difficult to achieve such a high level of security," explains Zimmerman. The pandemic and its repercussions on air travel also prompted the company to stop an experiment started in summer 2019, which offered guaranteed fares for certain specific itineraries by reimbursing passengers if the price dropped before take-off. According to Zimmerman, the project could restart soon, when the sector starts to stabilize.
Giorgos Zacharia, president of the online travel agency and flight search engine Kayak, says the company employs a team of PhD students from the Massachusetts Institute of Technology (MIT) who are dedicated exclusively to refining the site's price prediction tool. While the company's forecasting algorithm, introduced for the first time in 2013, usually had to be adjusted every couple of years, in the last two it has undergone "major retraining" after a few months, and sometimes weeks, from each other. Zacharia adds that the accuracy of forecasting tools, which typically hovers around 85 percent, may have dropped periodically in recent years, possibly approaching more than 83 percent. This means that, at certain times, waiting or buying a ticket when the company's site recommended doing so was less likely to lead to the lowest possible price.
"Machine learning loves learning from repeatable old patterns and make predictions based on the probability that these models will work again - explains Zacharia -. So the pandemic, which brings with it many unexpected anomalous events, also affects the input data of models like this ".
Hayley Berg, Hopper's chief economist, reports that the company's predictive tool has been trained with 75 trillion routes and eight years of historical price data. Today, however, the algorithm takes greater account of what it has seen over the past three years, something the company says has helped it maintain 95 percent accuracy throughout the duration of the pandemic. Even in the early days of Covid-19-related closures, Hopper hit the price predictions for airline tickets 90 percent of the time. Nonetheless, customers shouldn't be surprised by the volatility of prices: Hopper found that a domestic flight to the United States, for example, changes prices on average 17 times within two days, which becomes 12 in the case of an international flight. >
The hoaxes All of these changes favor the spread of many conspiracy theories among airline ticket buyers, including those who do not use price prediction platforms. The airline executives reiterate that obviously companies do not track user cookies, raising prices when they see that you are interested in a certain route (Zacharia, the president of Kayak, says that fares are occasionally higher or lower depending on location at the time of the search, because the systems take into account the "point of sale"). And there is also no reason why flights should be cheaper on Tuesdays than any other day of the week, as is often heard among bargain hunters: “The best time to book depends on the trip, in particular. from the origin, the destination, and the time of departure and return ", explains Berg.
However, at the moment you don't need a sophisticated machine learning algorithm to understand when is the best time to buy a flight: there is no better time. The prices are so high, explains Victoria Hart, a spokesperson for Kayak, that there aren't "many indicators to suggest waiting".
This article originally appeared on sportsgaming.win US.
