Starting and learning
Machine learning for good
Machine learning for social good predicts which water points need repair, maps crops and deforestation from satellite images, forecasts disease and food insecurity risk and makes public archives searchable. These projects work best with local partners who will act on the result.
How machine learning is used for social good
Forecasting floods
Models trained on rainfall, terrain and river records forecast floods days ahead, including in places with no river gauges. Forecasts help only when they reach people and agencies early enough to act.
Real exampleGlobal prediction of extreme floods (Nature, 2024)Google researchers published an AI model that forecasts extreme river floods, including in watersheds that have no streamflow gauges.
Mapping activity at sea
Deep learning detects vessels and offshore infrastructure in satellite images and matches them with tracking data. This shows where fishing and other activity happens even when ships are not broadcasting.
Real exampleSatellite mapping of industrial activity at sea (Nature, 2024)Global Fishing Watch and partners combined satellite imagery, vessel GPS data and deep learning, finding that 72 to 76 percent of industrial fishing vessels are not publicly tracked.
Monitoring wildlife
Camera traps produce millions of photos, most of them empty. Image classifiers filter blanks and suggest species, so conservation teams spend their time on analysis instead of sorting.
Real exampleWildlife InsightsWildlife Insights gives researchers AI models, trained with TensorFlow, that classify camera trap images and filter out blank images.
Mapping buildings and settlements
Segmentation models trace building outlines in satellite images where maps are incomplete. The maps support population estimates, planning and humanitarian response.
Real exampleGoogle Open BuildingsAn open dataset of building outlines derived from high-resolution satellite imagery by a deep learning model, focused on Africa and the Global South.
Can you learn machine learning for social good?
Yes. The technical skills are the same as in any machine learning work: Python, data handling, supervised learning and, for satellite and camera data, deep learning. What sets the work apart is partnering with the people who will use the result and evaluating it honestly in their setting.
- Python and data handlingLoad and clean public data such as survey tables, sensor records and open datasets.Learn it in Data Tools
- Statistics and uncertaintySampling, bias and uncertainty, which matter when data under-represents the places you care about.Learn it in Statistics
- Supervised learningClassification and regression with splits by region or time, so results hold in new places.Learn it in Machine Learning
- Deep learning for imagesConvolutional networks for satellite and camera trap images.Learn it in Deep Learning
Moving into machine learning for social good
If you already work in the field
People in development, conservation, public health or humanitarian work already know the problems and the partners. Add Python and machine learning basics, then start with a public dataset in your area and a question your organisation would act on.
If you come from data science or ML
Data scientists moving into social good need to learn how decisions are made on the ground, the limits of the available data and the ethics of working with vulnerable groups. Volunteer with an established organisation and agree on what success means before building a model.
Roles that do this workData scientistGeospatial data scientistMachine learning engineerMonitoring and evaluation specialist
Tools you will use
| Tool | What you use it for |
|---|---|
| pandas | Cleans and joins public data from government and NGO sources. |
| scikit-learn | Trains classification and regression models for risk and need prediction. |
| GeoPandas | Joins records to districts and maps, so predictions can be analysed by place. |
| Google Earth Engine | Analyses satellite imagery for crop, water and land-cover monitoring; free for academic and research use. |
| QGIS | Inspects and maps geospatial data and model outputs with a free desktop GIS. |
Public data to practise on
- Pump it Up: Data Mining the Water Table: Water-point records from Tanzania labelled as functional, needing repair or not working; free registration on DrivenData.
- World Bank Open Data: Country-level development indicators on health, education, poverty and environment for analysis and forecasting.
Your first project
Download the Pump it Up water-point data from DrivenData, predict whether each water point is functional with a random forest, and compare it with a majority-class baseline using accuracy and per-class recall.
Common questions
What is AI for social good?
Using machine learning and related methods for problems such as disaster response, conservation, health access and poverty mapping. The work is usually done with non-profits, governments or researchers who act on the results.
How can I volunteer my machine learning skills?
Look for data volunteering programmes, open-source projects from conservation and humanitarian groups, and research groups that need help. Start with a scoped project where a partner has agreed to use the result.
What are the risks of machine learning for social good?
Models trained on data from some places can fail in others, and predictions about people can cause harm if used without consent or oversight. Work with local partners, test where the model will be used and be clear about its limits.
Courses for these skills
- Practical Machine Learning in PythonLearn regression and grouped evaluation.51 published lessons · more lessons in production
- Deep Learning with PyTorchStudy image representations and visual recognition.Syllabus available · lessons open as produced
- Data Tools: NumPy, pandas, Matplotlib and scikit-learnKeep catalogue identifiers and extraction metadata traceable.23 published lessons · all planned lessons open
- Statistics and Probability for Machine LearningAssess uncertainty and sampling limitations.41 published lessons · all planned lessons open
- Generative AI: LLMs, RAG and AgentsConnect document retrieval with source-grounded answers.Syllabus available · lessons open as produced
New to programming or maths? Start with Python, or follow the machine learning path in order. Published lessons are free to read; compare Free and Pro.