A Machine Learning Based Approach for Prediction of Actual Landing Time of Scheduled Flights

Abstract

In India, the Central Air traffic Flow Management system (C-ATFM), New Delhi has introduced the Ground Delay (GD) program to manage the airport capacity constraints of three metro airports (New Delhi, Mumbai and Bengaluru). The predictability of arrival(landing time) has a direct impact on the GD program, tactical air traffic control (especially Conflict Detection, Flight level allocation) and Sector capacity. The predictive analysis study used in this paper uses a range of statistical techniques from supervised machine learning and data mining technique on historical data to make predictions for landing time from departure information. The predictive modelling developed using Multi Linear Regression (MLR) model proposed in this study can lead to an accurate prediction of the actual landing time at the time of departure using minimum attributes. Exponential moving average of historical flying time traces non-stationary flight time variation. The proposed MLR model gives lesser Root Mean Square Error (RMSE) for predicting landing time comparing to Estimated Landing Time (ELDT) prediction by existing ATM (Air Traffic Management) automation system. In addition to the highlighted significant factors, the study gives an insight into the root cause for the early arrival of the aircraft and congestion at the capacity-constrained airport, which is one of the perpetual problems of Indian air traffic management. This analysis uses data collected by C-ATFM Delhi, India, during the period of October–November 2018.