Inspiration
The increasing frequency and severity of natural disasters around the world have highlighted the urgent need for improved predictive models. Inspired by the potential of artificial intelligence to transform disaster preparedness and response, we embarked on a project to create an AI model capable of predicting the occurrence of future natural disasters. By leveraging data from NASA's Earth Observing System Data and Information System (EOSDIS), we aimed to develop a tool that could help regions prepare more effectively and mitigate the negative impacts of these events.
What it does
It's a machine learning model that can be used to forecast the occurence of natural disasters, we deployed the model inside a web application with the main objective of it being accesible to the general public and providing accurate predictions of the likelihood of natural disasters such as floods and storms.
How we built it
- Data Acquisition: We started by acquiring relevant data from EOSDIS, focusing on the historical data of previous natural disasters.
- Exploratory Data Analysis: We conducteda EDA to understand the underlying patterns and correlations of the different features on the dataset, this step allowed us to take a closer look as to what key features and relationships could be used to improve the accuracy of the model
- Data Preprocessing: The data was cleaned and normalized as to make it suitable for the model. This mainly involved handling missing values and leaving out features that could be irrelevant to the predictions.
- Model Training and Evaluation: The model was trained with the data acquired and evaluated through its corresponding metrics.
- Deployment: The model was deployed as a web application, providing with users with predictions of potential natural disasters in their region.
Challenges we ran into
- One of the major challenges we faced was the little knowledge we previously had about what could cause natural disasters and what features could be used to efficiently generate predictions, so we had to research a lot to have a general idea of what could be appropiate to use.
- Another challenge we faced was ensuing the data quality was appropiate for our project, for this we had to look into several sources and datasets until we found the right one that fit the characteristics we were looking for.
Accomplishments that we're proud of
- High-Accuracy Predictions: We achieved significantly accurate results that were coherent with the trends that the historical data showed.
- User-friendly Application: Developed an intuitive web application that provides predictions making it accesible to the general public that may be beginners on the subject but want to have an insight on the forecast of these occurences.
What we learned
- With a background in computer science, we delved into the patterns and characteristics of various natural disasters. Learning about the environmental triggers and precursors to events like hurricanes, earthquakes, and floods.
- Through our research, we gained a deeper understanding of how climate change influences the frequency and severity of natural disasters.
- Applying computer science principles to real-world problems showed us the practical implications of our work. Understanding how our model could help in disaster preparedness and response was a powerful motivator and a significant learning outcome.
What's next for Early Detection of Natural Disasters
We have to continuosly refine and enhance the model by incorporating new data and feedback from users and experts on the subject alike as to help maintain and improve predictions over time and overall add more valuable information into the project. We could develop and integrate real-time monitoring of events as to have a comparation between what's currently happening and what could happen according to the model's predictions. Also, we could continue research and development effort to explore other machine learning techniques and techonologies that could further enhance the quality of the project.
Built With
- blazor
- python
- tensorflow
Log in or sign up for Devpost to join the conversation.