Predictive analytics for Flight price prediction using mix of regression and ensemble modelling techniques
For the predictive analytics problem of airfare prediction,different supervised machine learning regression techniques like Multiple regression, Regularization( Ridge, Lasso, Elastic net), Decision tree, ensemble methods like Random forest along with boosting ( GBM, XGboost) were used. We looked at over fitting and under fitting and variable transformation along with feature scaling and selection. Different hypothesis were generated like flight price on weekend and price on weekday is more costly than weekend and night time respectively.
The project highlighted use of predictive analytics in Airline industry where price is very dynamic depending on type of airline, time of day and day of week etc. We concluded that predictive analytics can be a right forecaster of flight price based on historic pricing and give competitive advantage to airline industries. Use of ensemble techniques along with hyper parameter tuning ensured that model has good accuracy and also reduced the problem of overfitting for new unseen test data. Feature scaling and selection plays very important role in part of model building
Tools and Techniques Used
- Anova
- Decision Trees
- Random Forest
- R
- Tableau
- Ensemble
- Machine Learning
- Linear Regression
- Data visualization and EDA
- Cart and KNN
- Hypothesis Testing