Intended audience: DataKind Volunteers
At DataKind, our working definition of data science is using computers to interpret data, model relationships in data, and create data-driven algorithms. This often involves the use of advanced analytical and computational techniques to extract new information from novel data sources, usually for the sake of supporting organizational decision making or increasing efficiency. To sum it up then, data science is the necessary processing of otherwise overwhelming and messy information to help our human brains make better decisions.
So, what kinds of projects count as data science at DataKind? What is in and out of scope for a DataKind project?
It depends. We support organizations at different levels of data maturity, but data science solutions generally include at least one of the following:
- Collecting and merging external data sources (from government, social media, or other sources)
- Parsing or scraping difficult data sources (such as from images, text, or websites)
- Modeling or predicting the future
- Integrating the data project results to make better decisions, instead of just reporting on output into to an unchanging box
- Automating laborious tasks
Okay, so I’ve decided on a methodology, how detailed do I need to get in explaining and parsing out that methodology in the Design Stage?
At this stage, we limit defining the technical methodology to outlining the high-level approach for executing the project, rather than diving into details like exactly which model or software you will use. For example, you might make a plan like “We’ll first conduct exploratory data analysis (EDA) to identify trends, then develop a predictive model for homeless shelter demand based on weather forecasts. Finally, we’ll refine our model using real-world data feedback.”
In traditional data science projects, established methodologies often look like:
- Exploratory Data Analysis: Map the data, visualize the data, look for outliers, etc.
- Data Cleaning: Manipulate the data for use based on what you discovered
- Prototyping: Make a Minimum Viable Product (MVP) that allows the users to test the functionality of the product idea
- Executing: Adjust your prototype based on testing and finalize the model you have selected and/or the user experience for interacting with the data product you have created
- Repeat (as needed)
This might be the perfect method for your project, or you might have much more specific details ready to be outlined based on the results of the data audit. Your project might require a completely different framework, and that’s fine too! DataKind keeps the technical methodology flexible and open to innovation and creativity. Do what makes sense for the specific needs of your project, but whatever you decide, select a general method plan before you start the project.
Contributer(s): Benjamin Kinsella, Srivalya Elluru, Rachel Wells
Contact us
If you would like to learn more about us, partner with us, or get in touch, visit our website or email community@datakind.org.
Subscribe