South Sudan’s food security situation is assessed using the Integrated Food Security Phase Classification (IPC). Through this process, analysts periodically classify the severity of food insecurity in each county, ranging from Minimal to Catastrophe.

According to the most recent IPC analysis, approximately 7.7 million people ,57% of South Sudan’s population,were experiencing Crisis-or-worse food insecurity. The situation is driven by several interconnected factors, including conflict, low agricultural production, economic shocks and climate-related events such as flooding.

However, formal IPC classifications are conducted only a few times each year and require extensive analysis by food security experts. Between assessment cycles, humanitarian and government responders need faster ways to identify counties whose food security situation may be deteriorating.

The objective of this challenge is to develop a machine learning model that can provide an additional early-warning signal between formal IPC assessments.

Your task is to build a model that predicts whether a South Sudanese county will experience Crisis-or-worse food insecurity—IPC Phase 3 or above—during the current assessment period.

Your model will use information available before the period being predicted, including:

The county’s population

The year and month of the assessment period

The county’s IPC classification during its previous assessment

The percentage of the county’s population previously classified in IPC Phase 3 or above

The county’s most recently available cereal production figures

The county’s most recently available cereal production gap

This is a forecasting challenge. All features represent information that would have been available before the target assessment period. No current-period IPC information is included in the model inputs.

A successful solution could help humanitarian organisations and government agencies identify counties requiring closer assessment and prioritize limited field and response resources. The model is intended to complement the IPC’s expert-led assessment process—not replace it.

The winning solutions will be evaluated based on their predictive performance and their potential to support practical humanitarian decision-making.

Languages and tools:

You may only use open-source programming languages, packages and tools to build your solution.

Who can compete: This challenge is open to new users on Zindi who meet the challenge eligibility requirements.

Submission limits: You may make a maximum of 10 submissions per day and 300 submissions during the entire challenge.

Team size: The maximum team size is four participants.

Public-private split: Zindi maintains a public and private leaderboard for this challenge. The public leaderboard is based on approximately 30% of the test dataset. The private leaderboard is based on the remaining 70% and will determine the final rankings when the challenge closes.

Data sharing: The challenge data is available under a CC BY-SA 4.0 licence.

Platform abuse: Multiple accounts, sharing code or challenge information privately with participants outside your team, and other forms of platform abuse are prohibited and may result in disqualification.

Code review: Participants ranked in the top 10 on the private leaderboard will receive an email requesting their code after the challenge closes. Requested code must be submitted within 48 hours.

ENTRY INTO THIS CHALLENGE CONSTITUTES YOUR ACCEPTANCE OF THESE OFFICIAL CHALLENGE RULES.

Full Challenge Rules

This challenge is open to new users on Zindi who meet the challenge eligibility requirements.

Teams and collaboration

You may participate as an individual or as part of a team of up to four people.

When creating or merging a team, the team’s combined submission count must be less than or equal to the maximum number of submissions allowed at the time of formation. The newly formed team will receive the maximum number of submissions allowed for the challenge, minus the total number of submissions already made by its members.

Prizes are transferred only to individual participants or to the designated Team Leader.

Each participant may have only one Zindi account and may participate through only one individual account or team. Participants found to be using multiple accounts or collaborating across multiple teams will be disqualified.

Code must not be shared privately with participants outside your team. Any code shared outside a team must be made available to all challenge participants through the challenge discussion forum.

The Zindi data scientist who creates a team will be assigned as the default Team Leader. The Team Leader may transfer this role to another member of the team.

A Team Leader may invite other eligible data scientists to join the team. Invited participants may accept or reject the invitation. Until another participant accepts an invitation, the participant who created the team will remain listed as an individual on the leaderboard.

No new members may be added to a team during the final five days of the challenge or during the final hour of a hackathon.

A Team Leader may initiate a merger with another team. Only the Team Leader of the invited team may accept the request. Unless changed after the merger, the Team Leader of the initiating team will remain the Team Leader. Teams may merge only if the combined number of members does not exceed four.

A team may be disbanded if it has not made a submission. Once a team has made a submission, individual members may not leave the team.

All team members will receive the points associated with the team’s final ranking. Points are not divided among team members.

Datasets and packages

Solutions must use publicly available, open-source packages and tools.

You may use only the datasets provided for this challenge. External datasets are not permitted.

Automated machine learning tools, including AutoML platforms and packages that automatically generate complete modelling pipelines, are not permitted.

You may use pretrained models provided that they are openly and publicly available to all participants.

Challenge data may be accessed, used and shared for commercial, non-commercial, research or educational purposes under the Creative Commons Attribution-ShareAlike 4.0 International licence—CC BY-SA 4.0.

You must notify Zindi immediately if you become aware of any unauthorised access to or transmission of the challenge data. You must also cooperate with Zindi in addressing the unauthorised access or transmission.

Your solution must not infringe the rights of any third party. You must be legally entitled to submit your solution and, where applicable, assign the required rights in the winning solution code to Zindi.

Submissions and winning

You may make a maximum of 10 submissions per day.

You may make a maximum of 300 submissions during the entire challenge.

Before the challenge closes, you must select two submissions to be evaluated on the private leaderboard. If you do not select two submissions, your two highest-scoring public leaderboard submissions will automatically be selected.

During the challenge, your best public leaderboard score will be displayed, regardless of the submissions you have selected for final evaluation. When the challenge closes, your higher private leaderboard score from the two selected submissions will determine your final ranking.

Zindi maintains a public leaderboard and a private leaderboard for this challenge. The public leaderboard is based on approximately 30% of the test dataset. While the challenge is open, participants will be ranked according to the AUC score achieved by their submissions on this portion of the test set.

When the challenge closes, the private leaderboard—based on the remaining 70% of the test dataset—will be revealed and used to determine the final rankings.

A submission will only be counted after it has successfully completed processing. A submission that fails during processing will not receive a score and will not count towards the daily submission limit. Participants experiencing submission errors should request assistance through the challenge discussion forum.

If resource restrictions are specified on the challenge page, all participants must comply with them.

Participants ranked in the top 10 when the leaderboard closes will receive an email requesting their code. Requested code must be submitted within 48 hours and must follow the Reproducibility of submitted code requirements below. Failure to submit the requested code within the stated period may result in disqualification.

In addition to submitting reproducible code, finalists may be required to provide:

Brief documentation explaining their modelling approach

A one-page write-up covering the features used, validation strategy and practical interpretation of the predictions

The selected decision threshold and the resulting F1 score, precision and recall

A presentation during the IndabaX South Sudan 2026 Showcase

If your solution places first, second or third on the final leaderboard, you will be required to submit your winning solution code for verification. You agree to assign to Zindi the applicable worldwide copyright rights in the submitted winning solution, subject to the official challenge terms.

If two submissions receive identical final scores, the submission made first will receive the higher ranking.

Winners will be paid through bank transfer, PayPal for payments equal to or below USD 100, or another available international money-transfer platform.

International transfer fees will ordinarily be deducted from the prize amount. Where the prize is below USD 500, Zindi will cover the international transfer fees. Winners remain responsible for any fees charged by their own bank or receiving institution and for any applicable taxes.

Winning individuals or Team Leaders may be required to provide proof of identity, proof of residence and a bank letter confirming their banking details. Payments will be made in US dollars or the stated currency of the challenge. If a winner’s account cannot receive the relevant currency, the winner’s bank must provide confirmation, after which Zindi will attempt to arrange an alternative payment method.

Due to restrictions associated with the ongoing Russia–Ukraine conflict, Zindi may currently be unable to make prize payments to winners located in Russia. Any related cases will be handled individually.

Prize payments will be made after the completion of code review and the sealing of the final leaderboard.

Zindi may remove or disqualify an individual, team, account or submission if it reasonably believes that the participant has violated these rules.

Zindi is committed to providing solutions of practical value to its clients and partners. Zindi therefore reserves the right to disqualify submissions that compromise the usability, integrity or intended value of the solution. This includes the use of data leakage, prohibited data or any practice that undermines the forecasting nature of this challenge.

Because this is a forecasting challenge, participants must ensure that their models use only information that would have been available before the IPC period being predicted. The use of current-period or future information, directly or indirectly, will be treated as data leakage and may result in disqualification.

Zindi also reserves the right to disqualify participants or submissions that violate the rules, the spirit of the challenge or the platform’s Terms of Use, regardless of leaderboard position.

Please consult the challenge FAQs and Zindi’s Terms of Use for any additional rules. Zindi reserves the right to update these rules when necessary.

Reproducibility of submitted code

Your submitted code must reproduce the submitted predictions and leaderboard score. If it produces a different score, Zindi reserves the right to adjust your ranking to match the score reproduced by the submitted code.

If your code does not run successfully, you may be removed from the top 10.

You must set all relevant random seeds. Repeated runs of your code should produce consistent predictions and approximately the same leaderboard position.

If randomness causes your reproduced score to differ materially from your submitted score, Zindi reserves the right to adjust your ranking to the closest reproducible score.

Your code must train the model and generate the final submission file without requiring manual intervention.

The submitted code must use only the competition datasets and permitted openly available pretrained models.

Custom or private packages that are not publicly accessible will not be accepted.

The code must not use current-period target information, future IPC classifications or any other information unavailable at the time of prediction.

All required package versions, dependencies and execution instructions must be clearly documented.

The evaluation metric for this challenge is Area Under the Curve (AUC).

AUC measures how well your model distinguishes between counties that are at Crisis-or-worse food insecurity (IPC Phase 3+) and those that are not.

The target variable, food_insecurity_risk, is defined as:

1 = The county is in Crisis, Emergency, or Catastrophe (IPC Phase 3+)

0 = The county is below Crisis level (IPC Phase 1 or 2)

You will be awarded 3000 Zindi points for this challenge.