AltexSoft Creates Unique Data Science and Analytics-Based Fare Predictor Tool to Forecast Price Movements

Value Delivered

• 

  Uncovering the Hidden Patterns in Large Dataset

  To be able to predict the future price movements, our data scientists studied the historical data about airfare trend changes over the past several years. The actual information from millions of searches, processed by the Fareboom search engine was used, to make the predictions highly reliable. In order to increase the amount of data and statistical power, we handled neighbor travel dates data and clever merging algorithms to form the time series for further forecasting. Employing advanced data mining and aggregation techniques, the team was able to understand and visualize the hidden patterns.
• 

  Predicting the Future Price Movements with High Confidence

  Based on discovered patterns, the team formed algorithm models with different parameters. The info about actual flight fares helped us find the most suitable prediction algorithm: Setting the system parameters of the recent past, our data scientist ran the algorithm to predict the fares we already knew. Thus, we were able to validate or disprove the hypothesis based on exact information. The final algorithm has an average confidence rate of 75% and uses a time series forecasting to make both long-term (7 weeks) and short-term (7 days) predictions. The algorithm is constantly being improved through machine learning techniques, based on the factual information about the confirmed and disproved predictions.
• 

  Seamless Price Predictor Integration and Optimized Experience

  The Price Predictor feature is integrated into the existing fare search functionality and is shown to a certain segment of the Fareboom users (currently about 20%). Being displayed in a form of a search module and a popup window, it is sure to grasp the user’s attention and allows for a multitude of interactions (e.g. Close Popup, See More) and events (e.g. scroll, hover over, etc.). Therefore, we can track multiple stats and adjust the user experience based on them. For example, we have found that Price Predictor has doubled the average time per session within a month since the release and continues to grow the conversion.