Forecasting Dengue Outbreaks in Sri Lanka
CO425: Final Year Project II
Group 20
Team Composition
Supervisors
Dr. Namal Karunarathne
Department of Computer
Engineering,
Faculty of Engineering, Peradeniya
Prof. Faseeha Noordeen
Department of Microbiology,
Faculty of Medicine, Peradeniya
Mr. Imesh Ekanayake
PhD Candidate,
RMIT University,
Australia
Team members
Lakshitha Konara
E/18/181
Ravindu Mihiranga
E/18/224
Hirusha Uthsara
E/18/368
Agenda
1
Problem Statement & Solution
2
Methodology
3
Model Selection & Training
4
Findings
5
Deliverables & their Impact
6
Demonstration
Problem
Statement
Dengue Outbreaks in Sri Lanka
Dengue Outbreak in 2017
  • A total of 186,101 suspected cases
  • 440 dengue-related deaths occurred.
Consequences of Absent Dengue Outbreak Prediction System
1
Delayed Response and Increased Transmission
Without a prediction system, response to dengue outbreaks is delayed, leading to higher transmission rates and severity of cases.
2
Strained Healthcare Resources
Absence of prediction systems strains healthcare resources, potentially overwhelming facilities due to inadequate preparation for managing dengue cases.
3
Economic and Social Impact
Communities face increased economic costs due to healthcare expenditures and lost productivity, along with social disruption and public panic during outbreaks.
What if? …
A solution where we can predict dengue outbreaks.
Solution Overview
Providing an early warning system to help manage potential dengue outbreaks in Sri Lanka using environmental factors.
Data Analysis
Explanatory data analysis is used to identify patterns, trends, and relationships within the data.
Predictive Modelling
Data preprocessing.
Model selection and training.
Model evaluating.
Early Warning System
Creating a web interface to forecast dengue cases in a given month of the year.
Methodology conducts to solve the problem
Data Collection
Collected data:
1
Monthly Weather Parameters from 2007 onwards: Precipitation, Temperature Min & Max, Humidity, and Wind data for two-time stamps
Department of Meteorology, Sri Lanka
2
Weekly Dengue Patient Count from 2007 onwards
Epidemiological Unit, Ministry of Health, Sri Lanka
Finalized Districts
  • Colombo
  • Gampaha
  • Kandy
  • Galle
  • Kurunegala
  • Jaffna
  • Puttalam
  • Rathnapura
  • Batticaloa
Data Preprocessing
  • Patient Data Preprocessing
  • Convert weekly data into monthly data.
  • Remove patient data of unwanted districts.
  • Weather Data Preprocessing
  • Combine collected weather data together
  • Check for data types & units
  • Handle Imputations using KNN Clustering
  • Using nearest data points to fill the missing data.
  • Appropriate for weather parameters to be contextually consistent with observed data patterns.
  • Feature Scaling
Explanatory Data Analysis
Total Patient Count from 2007 to 2023
Total Patient Count with the Districts from 2007 to 2023
EDA Finding 1: Colombo & Gampaha Comparison
  • Able to find some resemblances between the patient count recorded and the weather parameters.
EDA Finding 2: Kandy & Kurunegala Comparison
  • Found the same kind of resemblances, like between Colombo and Gampaha.
EDA Finding 3: Time Lagged Correlation Analysis
  • Theory: Effects of weather parameters like precipitation and humidity have a lagged impact on behavior, and also the confirmation of a new patient.
  • For example, female mosquitoes (Aedes aegypti) lay eggs 10 days after rain, eggs mature in 10 days, and laboratory confirmation of a new patient after infection, takes up to 14-16 days.
  • Mature female mosquitoes can live actively transmitting dengue disease for up to 56–60 days, allowing prolonged risk of disease spread.
Project Workflow
Models Selection & Training
Considering the results of EDA we decided to train the below models for each district separately.
1
SARIMAX (statistical)
  • Capacity to manage seasonality and exogenous variables in time series analysis.
2
LSTM
  • Ability to capture long-term dependencies and temporal patterns in the data.
3
CNN
  • Ability to identify spatial and temporal patterns in the data.
4
XGBOOST
  • Capture complex historical trends and environmental influences using lagged values.
Demonstration
Deliverables and their Impact
1
Public Health Impact
Beneficial for proactive public health measures, minimizing the impact on communities.
2
Resource Allocation
Efficient allocation of resources, aiding in the management and treatment of dengue-affected individuals.
3
Epidemiological Insights
Forecasting can offer crucial insights into the patterns and spread of the dengue virus, aiding in epidemiological research and control.
Findings
1
A relationship between weather parameters and patient count
2
Found a similar pattern between weather parameters and patient count in two districts
Ex: - Colombo & Gampaha, Kandy & Kurunegala
3
Found a time lagged relationship between patient count and weather parameters
4
Found the most accurate ML model for this scenario after trying out various models
Identified Areas for Enhancement
  • Low sensitivity of weather parameters and a smaller number of data points.
  • May have effects from other factors like population density and garbage disposal which are not available.
  • accuracy and relevance of weather data (Katunayake station data considered as Gampaha district)
Thank You!
We appreciate your time and attention.
Q & A
Made with