Passenger flow prediction : Finding and developing a sustainable machine learning model for airport passenger flow prediction

There are many outdated routines and processes in today's aviation industry that major airlines lack the motivation to update. While this may not hold any direct security concerns, it creates bottlenecks at checks and high salary costs for otiose airport personnel. This study aims to together with the company Objective Solutions examine the possibility to increase the costeffectiveness in the security checks at Arlanda, tested on terminal 5, using a machine learning model which would serve as the basis for the scheduling of personnel. When performing this study, appropriate model alternatives were identified based on model characteristics and the task given. Three models were extensively explored and developed, Seasonal Autoregressive Integrated Moving Average Exogenous (SARIMAX), Holt Winters Exponential Smoothing (HWES) and Long Short Term Memory (LSTM). These were tested using real data collected from the airport database obtained through SQL. The model was built using Python in the Google Colab platform, the data was first handled and restructured and was then run through the different models with equal prerequisites. The models were evaluated using three different measuring tools; Mean Squared Error (MSE), Mean Absolute Error (MAE) and graphically. One of the models, Long Short Term Memory (LSTM), showed better accuracy than the others and was deemed successful in fulfilling the defined objectives of accurately identifying and predicting trends over the desired time period of two months. While this model was successful in reaching the defined requirements such as identifying trends and irregularities, the stochastic design of it entailed some instability which sometimes generated shifting results between runs, and it is up to Objective Solutions to decide if it is deemed appropriate to finalize the model into an end product ready for practical implementation