Business Unit: VisionWorks (Data Science)
Title: Forecasting video game sales
Technology: Python, R

Context & Concept:

There is an increasing demand in business for Robotic Process Automation. Companies are developing automated pipelines for predictive analytics in order to forecast sales, production output, operational expenses, etc. Such automated pipelines have the advantage of producing rapid results but come at the price of missing the subtleties of careful data modeling. In this internship you will develop an automated pipeline for time series forecasting and compare the advantages and limitations of an automated procedure to a more tailored approach to time series modelling. Finally, you will develop a dashboard/tool for time series analysis.

Goal:

You will use data from Kaggle on Video Games Sales:
https://www.kaggle.com/gregorut/videogamesales
You predict game sales for different sales markets, genres and platforms. You will develop an automated pipeline for time series forecasting and compare the results of the automated pipeline to the results of a non-automated procedure. Finally, you develop a dashboard for a company employee so he/she can make time series forecasts using a graphical user interface.

What we expect:
• You apply classic time series forecasting techniques (e.g. exponential smoothing models, ARIMA models) as well as machine learning techniques (e.g. random forests) to predict video game sales
• Forecasts can be disaggregated by market, genre and video game platform
• You show a deep understanding of time series modelling, rank models in terms of their performance, and justify your choice for selecting the best performing model(s) with the appropriate tests and metrics. Model choice explanation is of particular importance in time series forecasting. In practice you often get interrogated by business analysts who rely on ‘gut feeling’ for their forecasts and are often wary of predictions based on the dynamics found in the time series data.
• You evaluate whether model assumptions are met, and whether and to what extent possible violations impact the generalizability of your forecasts
• You show how you deal with outliers, incomplete data or erroneous data, and how they impact your results
• Any time series analysis comes with great visualization skills. They will help you get a feeling for your time series, and hint at possible difficulties and/or issues you will encounter during the modeling process;
• The final objective is to make an interactive dashboard (R shiny, Bokey, etc.) for a business employee. The dashboard should visualize the actuals and the forecasts, display the forecasts of the best performing model(s), etc. The app should be able to deal with outliers, data transformations, etc.

• Candidates need to be familiar with statistics, machine learning and basic programming. Knowledge of Python or R is recommended.
• You will work in our offices in Mechelen. Candidates need to arrange their own means of transportation.
• Candidates are able to work independently.