Skip to content

Reshiru/UNEP-STARTHACK22

 
 

Repository files navigation

START Hack 2022 - UNEP Case

Case Introduction

The core of our challenge is to design and shape the future of landslide prevention and management with the example of Hong Kong.

The important dataset related documents in this repository in the folder Dataset are named Case Description, Data Dictionary, Test.csv, Train.csv, and Sample_Submission.csv. Find the Zindi Platform here https://zindi.africa/hackathons/start-hack-22

Here you can find a great article from bbc why this is an urgent topic!

alt text

Help us detect real disasters in the future to save lives!


Hong Kong is challenged by landslides

Hong Kong, one of the hilliest and most densely populated cities in the world, is frequently hit by extreme rainfall and is therefore highly susceptible to rain-induced landslides. A landslide is the movement of masses of rock, debris, or earth down a slope and can result in significant loss of life and property. A high-quality landslide inventory is essential not only for landslide hazard and risk analysis but also for supporting agency decisions on landslide hazard mitigation and prevention.

As the common practice is visual, labour-intensive inspection, this hack focuses on automating landslide identification using artificial intelligence techniques and embedding this solution into the creative vision: “Harnessing the power of modern technologies to address landslide-management in Hong Kong”


Expected final product

During the hack you can submit unlimited coding submissions with real-time-ranking on the zindi platform, but only the final one will count and has to be sent together with the pitch to dominik.bisslich@un.org before 25.03.2022 10:00 CET. We also ask you for two draft submissions on the 24th with the code and pitch, the first at 10:00 CET and 22:00 CET. It doesn't have to be perfect, we would like to track your progess, guide you and support you as best as possible.

Our vision is to rethink landslide management in Hong Kong and integrate and make sense of modern technologies. Not only is quantification important, but also an entrepreneurial and innovative approach. The final product is therefore a vision of landslide management presented in 7 minutes based on classified input parameters using machine learning/deep learning. After that we got a short 3 min Q&A-session with our experts.

Building on our shared vision, the first step is to analyse the provided dataset and perform binary classification of landslides. Which input parameters such as precipitation intensity or slopes are more likely to cause landslides? For this purpose, we provide one dataset for training and one for testing. The choice of approach is up to you, whether machine learning or deep learning. We want to identify the variables that make landslides possible and use them to identify risk areas. Because of the wide range of possible analyses and influences, this task can be performed by people of all experience levels, from novices to those with extensive coding training. Machine learning knowledge is desirable, but not essential. The evaluation of this task runs on a F1 score, the highest wins this task.

Data, however, has no meaning without context. It remains questionable how these can be used. To make our vision become true we also need creative entrepreneurs that have the ability to innovate and rethink without boundaries. For example, can other data be used, can networking with other emergency services be made possible, can this format be transferred to other countries or cities? We deliberately did not give a focus here, but simply addressed "Harnessing the power of modern technologies."

Landslide management is already a big issue in Hong Kong, but the use of AI can help to transform it. There are several research projects about that right now. The dataset was created by Hong Kong University of Science and Technology.


Case Pitch

You will see the Case Pitch live at the Stage around 6 PM on Wednesday, 23rd of March. We will upload the slides after that.


Deep Dive Slides - Find the slides in the media folder

The slides for the two deep dives, the technical one and the disaster-management related one, will be uploaded after the presentation on Wednesday, the 23rd of March at around 10 PM. However we strongly encourage you to attend these presentations to catch a glimpse of the team and the possibility to ask questions directly after.

Video from HKUST: https://www.dropbox.com/s/4zzm2zgouwerr7l/Technical%20Introduction_By%20Prof%20Limin%20Zhang%20of%20HKUST.mp4?dl=0


Further Information

We will use the zindi-platform to evaluate the coding part of the submission. There you can upload as many submissions as you want and will see a real-time ranking of your solution compared to others. This helps you to evaluate your solution and see if you should focus more on the technical or entrepreneurial side of the hack.


Resources

The data set is provided in a .csv file. The only hardware component required is a computer and a programming environment (Python, R, Stata...) or data analysis tools (Excel...). We strongly encourage to use Google Colab to to increase your own computing power.

The choice of the exact approach is left to the students. All required data and information will be published in this repository. For the presentation of your results you have a total of 7 minutes with a format of your choice. This can be a PowerPoint presentation, a role play, or something else entirely. After that there will be a short Q&A-session lasting about 3 minutes with the experts in Natural Disaster Management and the Hack mentors.


Judging Criteria

We will evaluate your solution based on the technical sophistication and entrepreneurial spirit surrounding it. One tip for evaluating your technical model is using zindi for real-time ranking and the current F1-Score of your submission. So you know whether to work more on the model or the creative solutions you are envisioning.

You will find the detailed criteria down here in the table:

Evaluation Criteria Description Weight [%]
Complexity and Technical Sophistication How good is the quantifiable solution of the hack? For the START Hack we use the error metric called F1 score. This score ranges from 0 (total failure) to 1 which is a perfect score indicating perfect precision and recall. Highest F1-Score will reach 100 % in this category, other teams are evaluated against a benchmark and baseline. 60
Feasibility and Usability What steps are necessary to make this solution a reality? Here we look at the reproducibility of other situations and how users without a deep technical background can adapt these solutions. 10
Potential in harnessing modern technologies Modern technologies are diverse and also include IoT-enabled devices or drones? Are other technologies meaningfully integrated into the context and can they provide further benefits for landslide management? 10
Creativity, Innovation and Vision How original are these ideas and how could they revolutionize landslide management in Hong Kong? A holistic approach with the consideration of many stakeholders, applications and a look into the future is required!? 10
Delivered Pitch and Q&A How successful is the team's presentation? Is it clear what the team wanted to achieve and do the other evaluation criteria play in the same direction? Is the team spirit recognizable? 10

Meet your Hack Mentors here, just click on the picure to open the video :D

Meet Darius (Data Science expert from zindi) here:

IMAGE ALT TEXT HERE

Meet Melissa (graduate student in Computer Science) here:

IMAGE ALT TEXT HERE

Meet Maxime (graduate student in Robotics) here:

IMAGE ALT TEXT HERE

Meet Dominik (graduate student in Mechanical Engineering and Business Administration) here:

IMAGE ALT TEXT HERE

[online] Haojie (PhD Geoscience at Hong Kong University of Science and Technology) via https://cehjwang.github.io/:

Deep-Dive by HKUST https://raw.githubusercontent.com/START-Hack/UNEP-STARTHACK22/main/Media/Technical%20Introduction_By%20Prof%20Limin%20Zhang%20of%20HKUST.mp4?token=GHSAT0AAAAAABSTH36SKHP7YTEDTZMCZSDEYRYVWQA

Meet our experts in Disaster Management, Law, Emergency Response, and Project Management here, just click on the picure to open the video :D

Meet Muralee (Head of the branch Crisis Management Branch at UNEP) here:

IMAGE ALT TEXT HERE

Meet Paula (Project Coordinator for Modern Technologies at UNEP) here:

IMAGE ALT TEXT HERE

Meet Arielle (Master degree in Technology Law) here:

IMAGE ALT TEXT HERE

[online] Meet Eike (graduate student in Environmental Policy and Law) here:

IMAGE ALT TEXT HERE

Meet *bhijith (LLB Law; Chess Start-Up Founder) here:

IMAGE ALT TEXT HERE

Meet Matheus (Master degree in International Law) here:

IMAGE ALT TEXT HERE

Meet Aman (graduate student in Geoscience) here:

IMAGE ALT TEXT HERE

Meet Thomas (undergraduate student in Scientific Computing and Data Analysis) here:

IMAGE ALT TEXT HERE


Mentor Availability

All our Hack Mentors except for Haojie will be on Discord and present during the Hack. In the first three hours of the Hack, before the sample submissions and the four final hours we will be available at our table at the hack.

You can always contact us or our experts. If someone is attending online, we stated that in the description. Otherwise we are happy to set up a meeting and guarantee for changing roles at our table so you have a chance to meet everyone.

Feel free to reach out!


Prize

The winning team will get four DJI Mini 2 drones with which landscapes or people are photographed from above. Perhaps there are also landslides included ;D

Since you always learn something in a challenge and it will be hard to pick a winner, the runner-up team will also get something. A hackathon always brings you to your limits, you cross them and this can go to your own health, we would like to promote your wellbeing after the hackathon. That's why we are adding four fitness watches. This is the Huawei Band 6, four of them in black. Enjoy and celebrate yourself.

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Jupyter Notebook 79.4%
  • Python 20.6%