Challenge Summary
Mass atrocities are large-scale attacks against innocent civilians, offending our morality and threatening global security. President Barack Obama has directed U.S. government agencies including the United States Agency for International Development (USAID) to develop methods to detect and prevent mass atrocities. As part of its response to the presidential directive, USAID is partnering with Humanity United and TopCoder to build a computational model for predicting atrocities at the sub-national level. We are challenging you to participate in this important humanitarian initiative. Your task in this contest is to seek out new sources of data that have the potential to be useful in an algorithm for predicting the occurrence of future mass atrocities.
Overview
A mass atrocity is systematic or otherwise large-scale violence against innocent civilians. In the rest of this contest specification, we shall use the single word "atrocity" as shorthand for "mass atrocity". For a more complete picture of what we mean by this term, consult the text of the presidential directive on atrocity prevention.
This data discovery challenge is the first in a series of contests that will build a computational model for predicting atrocities. Like models for predicting weather events or disease outbreaks, our model will identify statistical patterns in the past that led to an atrocity and then look for similar patterns in the present. Such a model can only be effective if we have sufficient kinds and quantities of data to discriminate between conditions that will probably lead to an atrocity and conditions that probably will not.
The most promising data sets from this contest will be used as input to subsequent contests in which we develop a practical algorithm for calculating the probability that an atrocity will occur at a given place and time.
Data requirements
We are asking you to find and describe a data set that has the potential to contain statistical indicators for the occurrence of an atrocity. Data can measure any social, economic, or physical value.
A data set can contain observations of one value or of several related values. The data may have been collected from one geographical area or from multiple areas. If you can show a connection between your data set and a record of atrocities in at least one area where the data was collected, it will make your submission stronger.
Atrocities can occur anywhere. In fact, many well-known historical instances took place in what is now called the developed world. In this challenge, however, we are focusing on countries classified by the World Bank as lower-income through upper-middle-income. Submitted data sets must track data relevant to those countries.
Sub-national area: Past models for atrocity prediction relied on country-level data. In this challenge, we are looking for data on a sub-national scale. The area in which the data was collected can be a region, province, district, town, or any other geographical unit that is smaller than a country.
Sub-yearly frequency: The data set must have been collected through observations that were made more than once a year. A higher frequency of observation (daily or weekly) is more valuable than a lower frequency of observation (monthly or quarterly).
Minimum of 30 observations: The data set must include at least 30 observations. In other words, the value or set of values that is measured by the data must have been observed at least 30 times.
To maximize your chances of winning a prize, you should aim for the more desirable end of the spectrum. A desirable data set covers a small area or areas, has a high frequency of observation, and contains a large number of observations.
In addition to satisfying the statistical requirements above, each data set that you submit must be freely available and clearly documented.
Freely available: For us to consider your data set, it must be freely available to download and use. Everyone must be permitted to use it, for any purpose and for any length of time, at no cost.
Documented: There must be a clear record of how the data was collected and by whom. You must review this record and ensure that it is available to us. We encourage you to assess the validity of the data-collection method in your submission.
Finally, we are seeking submissions that show novel yet relevant thinking about the problem of predicting atrocities through computational methods.
Innovative: USAID and Humanity United would like to discover a wide variety of data sets that have not been previously explored for the purpose of atrocity prediction. Data sets that reflect new thinking about atrocity prediction are preferred to those that are part of the standard literature.
Submission format
To submit your data, please download the submission template in the file format of your choice: Word or Open Document. For each data set that you wish to submit, please fill out one copy of the template file and submit it separately. Do not bundle multiple files into one submission, or else we won't be able to review your data sets properly.
The template includes instructions in gray text to guide you as you fill it out. You may delete the instructions and are encouraged to do so. The first three parts of the template call for brief answers without subjective content. The fourth and final part is a free-form section in which you are invited to argue the merits of your data set in as much detail as you wish.
When you fill out the template, please do not include any personally identifying information. We want the evaluators to conduct a blind review, i.e., without regard for the identities or credentials of the contestants. Your TopCoder member profile already includes your contact information so that we can reach you in the event that you win a prize. Do not include your name, title, location, email address, or any other personal information in the submission.
Target audience
A panel of expert evaluators, some specializing in data analysis and others in humanitarian policy, will review each submission that passes a preliminary screening performed by the copilot. You are writing your submission primarily for the panel. A secondary audience to keep in mind is the community of algorithm competitors who will seek to validate the most promising data sets in subsequent contests.
Judging criteria
Submissions will be evaluated on the strength of their innovation, relevance, data quality, and overall quality. In the event that the same data set is proposed by several submissions, the one offering the stronger justification will be preferred.
Milestone phase
To receive feedback from evaluators prior to the final deadline, upload your submissions by the milestone deadline of 10:00 am EDT on Wednesday, March 13. Your milestone submissions will be screened by the copilot and forwarded to evaluators. You will receive feedback on Monday, March 18. In addition, up to five outstanding milestone submissions will be awarded $100 each. After the milestone feedback is delivered, all contestants will have seven days to complete their final submissions.
What to submit
Submission zip file: Fill out one copy of the submission template in the file format of your choice (Word or Open Document).
Source zip file: Identical to the submission zip file.
Preview image: Use the preview image supplied for this contest.
Final fixes
You may be asked to complete one round of minor changes to ensure that your submission meets the stated requirements of this contest. More information about final fixes.
Please read the challenge specification carefully and watch the forums for any questions or feedback concerning this challenge. It is important that you monitor any updates provided by the client or Studio Admins in the forums. Please post any questions you might have for the client in the forums.