Capgemini Data Science Take-Home (Choose One Exercise)

Overview

You have been asked to prepare a 20- to 30-minute presentation on a data science topic of your choice (selected from the options below). The purpose of this exercise is to demonstrate your ability to draw insights from data, present those insights in a business-friendly format, and confirm your coding knowledge. These topics are similar in nature to real client projects. Please ensure your presentation is accessible to a general technical audience.

---

Choose One Exercise

Select one of the following datasets and problem statements:

Option 1 — Movie Revenue Prediction

Your client is a movie studio that needs to predict movie revenue in order to greenlight projects and assign budgets. Most of the data is comprised of categorical variables. While the budget for a movie is known in the dataset, it is often an unknown variable during the greenlighting process.

Data provided: Movie_Revenue_Predictions.csv

The dataset includes fields such as title, tagline, revenue, budget, genres, keywords, original language, overview, production companies, production countries, release date, runtime, spoken languages, and status.

---

Option 2 — Article SPAM Classifier

News articles are sent to researchers to keep them up-to-date on technology. Lately, news feeds have been inundated with spam (pure advertisement) articles. The goal is to identify and eliminate spam articles from informative ones. How well can you classify articles as spam or valid?

The following sources have been pre-scraped, but you are welcome to scrape additional sources if needed to improve results:

• List of sites already blacklisted: article_spam_classifier__blacklist.json — a JSON array of domain names identified as spam sources.
• Valid news articles: article_spam_classifier__scraped_articles.json — a JSON file containing scraped valid news articles organized by outlet, each with link, publication date, title, article text, and author list.

---

Option 3 — Sales Forecasting

Build and evaluate models to predict national retail store sales for each store and department. Sales are highly seasonal, with key revenue periods around holidays such as Super Bowl, Labor Day, Thanksgiving, and Christmas.

Data provided: sales_forecasting.csv

The dataset contains weekly retail sales by Store and Department with an IsHoliday flag, spanning 2010–2012.

---

Deliverable

Prior to your interview, submit a single PDF file containing your full presentation (equivalent to a PowerPoint or Google Slides deck exported to PDF). Your PDF must include all of the following:

1. Description of the problem — state what you are solving/analyzing.
2. Presentation of insights and conclusions you generate from the data.
3. Relevant descriptive statistics (charts, graphs, etc.).
4. Specification of the predictive model (mathematical formulation).
5. Relevant model diagnostics.
6. Model interpretation — what do the coefficients or model outputs mean, and how would you use them?
7. Language, packages, and libraries used to develop your solution.

Your PDF must also include all code used to produce your analysis. The code must be executable (runnable) so that it can regenerate every result, chart, and table shown in the PDF.

---

Evaluation Criteria

• Presentation on analysis conducted that covers business outcomes and statistical methodologies
• Exploratory data analysis (EDA) to gather starting insights and conclusions
• Selection of ML/predictive modeling technique(s) and feature extraction
• Knowledge of data ingestion tools and languages
• Ability to conduct appropriate data cleansing
• Ability to code in open source languages

---

Rules

• Use any open source language of your choice.
• The solution you provide should be your own. You may reference any material you wish.
• Be prepared to discuss your code in depth during the interview (what it does, how it works, etc.).
• Utilize any statistical or ML technique(s) you deem relevant. For each technique you use, be prepared to discuss model diagnostics, results, and the mathematics behind your approach.
• There is no time limit on developing your solution.