Methodology

Data science

Powering decisions with machine learning and algorithms.

25 February 2021

IDinsight uses data science to support decision-makers to better predict, optimize, or model important program outcomes to amplify social impact.

About our data science work

Algorithms touch nearly every corner of modern life. These mathematical models dictate what shows up in our routine internet search results, determine which rideshare driver picks us up from a street corner, suggest how to fix a misspelled word in a text message, or even help grocery store managers optimally price a loaf of bread. There are equally ubiquitous opportunities to use data science to improve the efficacy and efficiency of the social sector.

The field of data science lies at the intersection of a number of domains, including computer science, statistics, and contextual knowledge (e.g. economics, physics, epidemiology, and more). At IDinsight we draw from these areas to provide three types of services to our partner organizations: (1) machine learning, (2) optimization, and (3) data modeling.

Our teams typically embed with country governments or NGOs so that we can support tailored support that is context-specific rather than delivering a “solution” with little support.

Our current projects

IDinsight’s data science team supports partners in a variety of ways. A few examples of our current work include:

Working with an education non-profit, to locate out-of-school girls. Using machine learning, we identified 250 percent more out-of-school girls per village than the organization’s standard enrollment campaigns. We project this will lead to the enrollment of 500,000 additional out-of-school girls over 5 years when fully scaled.
Helping a non-profit partner and national government deploy a chatbot that matches incoming questions about COVID-19 to a database of official public health messaging. We are now working to scale the solution to tens of thousands of users.
Supporting a national government to select optimal locations to build new school dormitories and run new bus routes to increase childrens’ access to school. To do this we have mapped the population density, road networks, costs and other constraints and then tested various algorithms to optimize allocations.

How we work

IDinsight’s data science team works in close partnership with other IDinsight teams to provide integrated end-to-end solutions for our clients. We support clients on their full journey from data collection → data systems design → deployment of cutting edge algorithms → evaluations of deployed solutions.

We work in close collaboration with our partners, striving to embed our team and our solutions within our client’s existing systems so that solutions are integrated into their day-to-day operations.

All solutions we build for our partners are tested for both bias and performance with careful documentation of our code that solutions are maintainable and extensible by IDinsight, our partners, and any future collaborators.

Machine Learning:

What questions can it help the social sector answer?

Where should we target my program to reach those most in need?
What is the most appropriate text response to citizens’ questions about COVID-19 regulations?
Which entitlement programs is this low-income household most likely eligible for?

What is it?

Machine learning learns relationships between variables in datasets. The methods are flexible to work with ordinary numeric data as well as text, audio files, and images.
Methods include: (un)supervised machine learning approaches, (Neural networks, random forests, k-means clustering, etc) that can be used for image recognition, remote sensing, natural language processing, and recommendation systems.

Optimization:

What questions can it help the social sector answer?

Where should we spend our infrastructure budget to increase education access subject to population location / density and travel time constraints?
Which subset of rural villages should we serve given limited personnel, budget, and travel time constraints?

What is it:

Optimization techniques help decision makers allocate limited resources as efficiently subject to a set of constraints. This includes solving problems around routing, task assignment, resource allocation and more.
Methods include: simulated annealing, genetic algorithms, linear programing, integer programming

Data Modelling:

What questions can it help the social sector answer?

Are COVID-19 cases likely to rise or fall in the next 2 weeks?
What might be the risk of COVID-19 spread through various modes of community health provision?

What is it:

Data scientists don’t always have mountains of data to analyze, but may be able to simulate complex systems to gain insights. E.g. An SEIR model might help the data scientist assess the rate at which a virus may spread under different social distancing conditions.
Methods include: Monte-Carlo simulations, network models, bayesian modeling, epidemiological modelings.