Abstract:
|
The objective of this Master Degree’s Thesis is to contribute to the Dynamic Pricing
Tool from EY’s Advisory & Advanced Analytics Department. The contribution
consists in two areas: Prediction and Optimization. A product portfolio’s history
of prices are used to compute demand predictive models for each product based
on prices, to further pass them through an optimization process to maximize the
revenue. By utilizing Machine Learning tools as Feature Engineering and stateof-the-art
prediction models such as XGBoost, an improvement of the Forecast
Accuracy is achieved. Then, by using the already computed predictive models,
Simulated Annealing and Broyden-Fletcher-Goldfarb-Shanno (BFGS) non-linearoptimization
techniques are applied, to find the best possible prices as input for
the previously computed non-parametric prediction models, with the objective of
maximizing the revenue. All these concepts are illustrated using a real client’s
dataset of beer sales history. The real product names were modified for confidential
purposes. |