FLIGHT FARE FORECASTING SYSTEM USING STACKED PREDICTION MODEL IN A WEB INTERFACE

Abstract

This project involves a web application
that predicts airline ticket prices using
advanced ensemble machine learning
techniques that incorporate features such
as departure date, airline, and
origin/destination locat ions. Using
historical records of air travel, the
application first cleans, pre processes, and
engineers features in order to evaluate
trends and relationships among various air
travel variables. After developing models
based on these relationships, the
ap plication uses a stacked predictive
model, as opposed to single analytical
model.
The stacked model combines two base
prediction models (Random Forest
Regressor and XGBoost Regressor) into a
single output generated from the meta
model (Linear Regression). The system
was evaluated by comparing Mean
Squared Error (MSE) and R squared (R2)
against each of the two base models. The
analysis of the feature importance
identified two statistically significant
variables in predicting air ticket price:
airline carrier and number of stops.