2. Olivier Van
Tongerloo
Data scientist
Duke Assam
Data scientist
Nico
Verheyden
Big Data
consultant
Wiljan Cools
Student big
data analytics
Wojciech
Kuberski
Student master
in artificial
intelligence
Kostas
Theodorakos
Software
developer
Ana Cristina
Rosa
Entrepreneur
Our team
The Juniors
DIHub bootcamp Graduates
4. Which features
will predict a
dengue outbreak
Why? When? Where?
Our challenge
Predict Dengue outbreaks using open data
5. Which features
will predict a
dengue outbreak
Why?
Enough time to
take measures to
prevent an
outbreak
When? Where?
Our challenge
Predict Dengue outbreaks using open data
6. Which features
will predict a
dengue outbreak
Why?
Enough time to
take measures to
prevent an
outbreak
Pinpoint a small
enough region to
be able to take
preventive actions
When? Where?
Our challenge
Predict Dengue outbreaks using open data
8. Our model can be extended with other data
and new regions
Scalable
Transferable
Reliable
iInformative
Our solution
General prediction model
9. No region specific data is used
Our model can be extended with other data
and new regions
Scalable
Transferable
Reliable
iInformative
Our solution
General prediction model
10. A prediction window of only 4 weeks is used
to be reliable but still usable
No region specific data is used
Our model can be extended with other data
and new regions
Scalable
Transferable
Reliable
iInformative
Our solution
General prediction model
11. Predictive features are returned by the model
A prediction window of only 4 weeks is used
to be reliable but still usable
No region specific data is used
Our model can be extended with other data
and new regions
Scalable
Transferable
Reliable
iInformative
Our solution
General prediction model
12. Binomial classification using time series
(Focus on Malaysia as most data concerning outbreaks was available from that region)
Our Approach
Technical solution
• Climate data (3 months)
• Confirmed dengue cases
• Data of neighbour province
• Demographics
Data preparation
• logistic regression
• Random forest (best)
Model building
• Test on test data
Model testing
14. • Add more (open) data to refine the model:
• Other countries
• Vegetation
• Lifestock
• Socio-economical data
• Logistics/transportation data
Next steps
Possible improvements
Scarce information on when, where and why a dengue outbreak happens
Preventive measures cannot be taken on a large scale
Preventive measures can be improved if we know why an outbreak happens
Scarce information on when, where and why a dengue outbreak happens
Preventive measures cannot be taken on a large scale
Preventive measures can be improved if we know why an outbreak happens
Scarce information on when, where and why a dengue outbreak happens
Preventive measures cannot be taken on a large scale
Preventive measures can be improved if we know why an outbreak happens
Scarce information on when, where and why a dengue outbreak happens
Preventive measures cannot be taken on a large scale
Preventive measures can be improved if we know why an outbreak happens
Model that is
Scalable
Transferable
Reliable (short term)
Informative
Model that is
Scalable
Transferable
Reliable (short term)
Informative
Model that is
Scalable
Transferable
Reliable (short term)
Informative
Model that is
Scalable
Transferable
Reliable (short term)
Informative
Model that is
Scalable
Transferable
Reliable (short term)
Informative