Prize

1st - $3000

2nd - $1500 

3rd - $1000

4th - $500

5th - $500

Challenge Overview

We are looking for predictive models to better replicate casual behavior of current social system state, predict future states, and modal alternate states resulting from intervention. 

Background

It's hurricane season in an area along the coastline. The population is diverse with distinct regions. The data provided span the first six hurricanes of a single hurricane season. 

Task Detail

 

We have some initial datasets collected as follows. The Census Table will represent the population at the start of the simulation, while the others will represent daily values spanning the simulation’s run time. In the latter tables, the “Timestep” field in each record specifies the day to which it pertains (or when the participant responded to the survey). 

 

Census Table: Demographics at the beginning of the hurricane season, broken down for each region or totaled over the whole area (the latter labeled as “All” in the Region column). The remaining columns are “Field”, “Value”, and “Count”, with the last being a tally of the number of residents in the region who have the given “Value” for the following fields:

• Population: Number of residents (including children). There is no “Value”, only a “Count”

• Gender: Number of region’s adult residents who are male vs. female

• Ethnicity: Number of region’s adult residents who are of each ethnic group

• Religion: Number of region’s adult residents who are of each religious group

• Age: Histogram of ages (in intervals of 5 years) across each region

• Employment: Number of region’s adult residents who do or do not have a full-time job (Value field is True or ���False, respectively)

 

Population Table: These are statistics aggregated over the entire urban area, broken down by “Timestep”.

• Deaths: cumulative number of deaths as of the given day

• Casualties: current number of residents who are either injured (requiring hospitalization) or dead on the given day

• Evacuated: number of residents who have left the area completely and have not returned by the given day

• Sheltered: number of residents at a public shelter on the given day 3

 

Regional Table: These are statistics for each day broken down by each region and day (“Timestep”).

• Deaths: cumulative number of region’s residents who have died this season as of the given day

• Casualties: number of region’s residents who are either injured (requiring hospitalization) or dead on the given day

• Sheltered: number of region’s residents at a public shelter on the given day.4

 

Actor-Pre Table: Actor-level survey, conducted when a hurricane is approaching:

Participant: ID for participant.

• Hurricane: Identifying number of hurricane (first hurricane of the season is 1).

• Demographics: gender, age, ethnicity (either majority or minority), religion (either majority, minority, or none), number of children, wealth (on a 0-5 scale, increasing from 0 to 5), pet owned or not, fulltime job or not, region of ���residence

• Survey questions:

  • At Shelter: Are you currently staying at a public shelter?

  • Evacuated: Are you currently residing outside the area?

  • Severity: The approaching hurricane poses a significant risk to the area. (Response on a 1-5 scale, ranging from “strongly disagree” to “strongly agree”) 

 

Actor-Post Table: Actor-level survey, conducted after a hurricane has passed:

Participant: ID for participant.

• Hurricane: Identifying number of hurricane (first hurricane of the season is 1).

• Demographics: gender, age, ethnicity (either majority or minority), religion (either majority, minority, or none), number of children, wealth (on a 0-5 scale, increasing from 0 to 5), pet owned or not, fulltime job or not, region of ���residence

• Survey questions:

  • At shelter: Did you stay at a public shelter at any point during the previous hurricane?

  • Evacuated: Did you evacuate the area at any point during the previous hurricane?

  • Injured: Did you suffer any injuries during the previous hurricane?

  • Risk: The previous hurricane posed a significant risk to myself and my family. (Response on a 1-5 scale, ranging from “strongly disagree” to “strongly agree”)

  • Dissatisfaction: The government response was fair and adequate. (Response on a 1-5 scale, ranging from “strongly disagree” to “strongly agree”)

 

Hurricane Table: Entries made for only those days (i.e., “Timestep”) when a hurricane is present:

• Hurricane: Identifying number of hurricane (first hurricane of the season is 1).

• Category: The category of the current hurricane (1-5 scale, with 5 being the most severe)

• Location: The region where the hurricane is projected to make landfall (if it has not yet landed) or where it is currently centered (if it has). If the hurricane has moved out of the area, the location will be given as “leaving”.

• Landed: Whether or not (“yes” or “no”) this hurricane has made landfall.  

 

There are two types of variables, stable variables and changeable variables . Stable variables are the one which do not change as time goes, while changeable variables are those which changes as time goes. These variables only apply to Actor-pre tables and actor-post tables. Only the following variables are changeable: 

• Time

• Severity/Risk

• Dissatisfaction

• Evacuated

• Shelter

• Injured

• Wealth

 

Final Predictive Goals: Given all these data of 6 hurricanes, we will need to build models to predict the situations for a new hurricane. You may want to build separate models for different questions, but please note that these multiple predictions are highly relevant to each other. That’s why they are included in the same challenge.

 

Short-term Predictions (Instances 3, 4, and 5)

1. Global Prediction

   1. How many people will be casualties (dying or suffering a serious injury) during the new hurricane?

   2. How many people will evacuate at least once during the new hurricane?

2. Local Prediction

   1. Which regions will suffer the highest/lowest percentage of casualties?

   2. Which regions will have the highest/lowest percentage of evacuations?

3. Individual Prediction

   1. Will TargetActor be injured or die during the new hurricane?

   2. Will TargetActor evacuate during the new hurricane?

    Long-term Predictions (Instances 6, 7, and 8)

1. Global Prediction

   1. How many people will die?

   2. How many people will evacuate at least once? 

2. Local Prediction

   1. Which regions will suffer the highest/lowest percentage of deaths? 

   2. Which regions will have the highest/lowest percentage of residents who evacuate at least once?

3. Individual Prediction

   1. Will TargetActor survive the following hurricane season?

   2. How many times will TargetActor evacuate during the following hurricane season

���Goal of This Challenge:  

You are asked to build models to make predictions to the corresponding questions. We will provide the answers for instances 3, 4, 6, and 7. Your solution will be judged based on the novelty as well as the performance on Instances 5 and 8.

The full dataset can be downloaded here.

Final Submission Guidelines

Submission

The final submission must include the following items.

• A write-up of your proposed model, i.e., how do you utilize the input to make the predictions.

• PoC solution is required.

• Predicted answers for instances 5 and 8.

 

Judging Criteria

Winners will be determined based on the following aspects:

• Model Effectiveness (50%)

  • Are your predictions on Instances 5 and 8 accurate?

• Model Feasibility (30%)

  • How easy to deploy your model?

  • Is your model’s training time-consuming?

  • How well your model/approach can be applied to other problems?

• Model Novelty (10%)

  • Are you using any novel model?

• Clarity of the Report (10%)

  • Do you explain your proposed method clearly?

 

Submission Guideline

You can submit at most TWO solutions but we encourage you to include your great solution and details as much as possible in a single submission.