# Choosing a Project ### Focus on Problems First, Then Data Rather than starting with a dataset and trying to find something interesting to do with it, identify a meaningful problem that could benefit from machine learning. This approach tends to lead to more impactful and realistic projects. ### Look for Human Pain Points Consider processes or tasks that are: * Time-consuming for humans * Tedious or repetitive * Prone to human error These often make excellent candidates for machine learning solutions. ### Project Sources #### Bring Your Own Project If you already have a problem in mind or access to interesting data: * Present your idea in the course or course chat * If you have any doubts about the project, formulate specific questions and ask in the course chat #### Partner with Organizations Local companies and academic departments often have real-world problems waiting for ML solutions: * Reach out to businesses or research groups in your area * Inquire about data-intensive problems they're facing * We can help you define project parameters with your partner * If data confidentiality is a concern, you can use the NDA template provided [here](#non-disclosure-agreement-nda) #### Public Datasets While convenient, using public datasets sometimes comes with limitations: * Projects tend to focus more on model optimization than real-world implementation * You miss valuable experience in data preparation and feature engineering * The problem may be artificially clean compared to real-world scenarios However, if you choose this route, consider adding complexity by: * Combining multiple datasets * Creating your own validation methodology * Adding constraints that reflect real-world conditions ### Data Resources If you're looking for public datasets, here are some valuable repositories: * [**ChallengeData**](https://challengedata.ens.fr/challenges/challenges_search): Datasets from real-world challenges * [**Hugging Face Datasets**](https://huggingface.co/datasets): Collections ready for NLP and other ML tasks * [**UCI Machine Learning Repository**](https://archive.ics.uci.edu/): Classic, well-documented datasets * [**Kaggle**](https://www.kaggle.com/datasets): Competitive datasets with community solutions * [**Roboflow Universe**](https://universe.roboflow.com/): Computer vision datasets * [**Papers with Code**](https://paperswithcode.com/datasets): Datasets linked to research papers ### Examples of Past Projects To get inspiration for a project, you might also want to review the projects nominated for the past VDE Machine Learning Prize in 2025 presented in the document below. {% file src="" %} ### Non-Disclosure Agreement (NDA) If you need an NDA for data you are getting from an organization or partner, you can use the following: {% file src="" %} Non-Disclosure Agreement (NDA) {% endfile %}