Case Study

DataKind- DataDive

 image: http://www.datakind.org/projects/predicting-and-preventing-human-rights-abuses

image: http://www.datakind.org/projects/predicting-and-preventing-human-rights-abuses

 

Project Overview


Implementing Team

Over 20 DataKind volunteers

Partner/Client:

Amnesty International

Resources

http://www.datakind.org/projects/predicting-and-preventing-human-rights-abuses

About

Amnesty International’s Urgent Action Network is a powerful program aimed at preventing potential human rights violations before they happen. After digitizing 25 years of data representing 5000 Urgent Action cases, the organization had1.4 million lines and 11,000 data files of human rights threats from 50 different countries. In November 2013, the organization participated in a weekend DataDive to see if DataKind’s volunteer data scientists couldhelp them create a predictive model that leveraged their existing dataset on past human rights threats to identify patterns of violence that could help predict and prevent future life-and death situations. 

Additionally, the team was also tasked with streamlining the process of coordination between three million human rights supporters when immediate action is needed.

Methods in Action


Approach

Open call

DataKind makes an open call for partner organizations to work with during the weekend DataDive.,

Selection of project partners
Amnesty international is chosen as one of the project partners for the weekend.. DataKind staff and volunteers work closely with the organization to understand its needs and scope a defined project.

Data Preparartion

The organization shares its data ahead of the event with DataKind staff and volunteer team leads called Data Ambassadors. They review and do preliminary data analysisto ensure the data is sufficient to answer the defined question. In some cases, DataKind staff may work to supplement the data provided by an organization with other sources.

Project Brief

DataKind staff and Data Ambassadors create a detailed Project Brief outlining the challenge, project objectives, datasets and key questions for the team to answer. This is shared with volunteers at the start of the DataDive.

Team Formation

At the start of the DataDive, each organization participating pitches their projects to the crowd and volunteers self-select to form a team that is led by one to two Data Ambassador team leads.

Pattern Identification

With the use of data mining techniques, the network of volunteers started to find patterns that predicted others. Subject matter experts from Amnesty International provided context for the volunteers and collaborated with them in exploring and understanding emerging patterns.

Prediction Modeling

After the patterns were identified, the volunteers were able to create a preliminary predictive model to identify high risk situations. In this case, the model identified words or terms in the Urgent Action alerts that were most often associated with a given threat escalating. Using this linguistic algorithm, Amnesty International could now predict which alerts were most at risk to escalate to a crisis point.

Process Infrastructure Redesign

While not part of the original task, the team of volunteers provided recommendations on the infrastructure and design changes for the Urgent Action alert process. Moreover, they also suggested changes to the format of emails and web page that included adding a map showing real-time data of actions around the world, among other internal changes.

Results

Product Output

At the end of the DataDive, DataKind provided Amnesty International an initial version of the prediction model that could be refined over time, with key terms to identify human rights threats at risk of escalating to serious cases.

In addition, DataKind provided a set of recommendation on how to incorporate more real-time data that can drive supporter engagement and awareness within Amnesty’s work, as well as data collection and storage best practices to improve data quality and future analytics.

 

Related