Page 84 - AI for Good Innovate for Impact

AI for Good Innovate for Impact

(continued)

Item Details
Model Training and The core model for this project is a Bayesian Hierarchical Model.
Fine-Tuning Additionally, it will leverage
• Power BI's built-in AI features (for initial insights, visualizations) [5].
• Python-based machine learning libraries (specifically Pandas and
GeoPandas for data handling and spatial operations, which are foun-
dational for applying ML models).
Testbeds or Pilot The deployment is not a separate phase after development. Implying
Deployments the pilot deployment is happening alongside development to gather
feedback early.

Code repositories Not Available

2 Use Case Description:

2�1 Description

This project harnesses the power of AI-driven geospatial analysis to pinpoint and visualise
clusters of otitis media and hearing loss within remote communities across Australia's Northern
Territory (NT). These clusters are then meticulously overlaid with crucial social determinants
such as housing conditions, access to healthcare, and educational attainment, offering a
comprehensive view of contributing factors. To ensure unparalleled spatial resolution and
granularity, acknowledging the NT's vast and diverse geography, the project utilises geocoded
community-level data provided by Telstra. This rich dataset comprises precise latitude and
longitude coordinates for 93 distinct NT communities, each enriched with metadata on
settlement type (e.g., village, minor town, major town, family outstation) and mobile phone
coverage [4]. This granular approach allows for the visualisation of each community as a discrete
spatial unit, avoiding reliance on broader, less precise aggregations like postal codes, thereby
preserving the fine-grained variations between communities. A custom seasonal classification
for the NT has been developed and directly implemented within Power BI using Data Analysis
Expressions (DAX) logic. This dynamic classification assigns each service event to a specific
season based on the month and health service region. Power BI's time series layers facilitate the
filtering and viewing of seasonal trends over time and across different regions, further enhanced
by the availability of time sliders. The ultimate aim is to create an interactive geospatial map
that provides actionable insights to inform health policy and interventions. The project's core
requirements include the creation of an AI-powered, interactive geospatial map that distinctly
showcases otitis media clusters within remote Northern Territory communities, overlaid with
pertinent socio-economic data. AI technology and analysis are central to generating ear disease
heatmaps and spatial clusters, enabling the identification of high-risk areas for otitis media
based on social determinants. Datasets concerning hearing loss and ear disease have been
acquired through de-identified health records and diagnostic assessments from health clinics.
It is a known limitation that this data is manually entered, which introduces a potential margin
of error. However, this is mitigated by annual audits conducted by the Continuous Quality
Improvement (CQI) manager, nursing team, and data entry officer. Summary public census
datasets for each community, specifically from the Australian Bureau of Statistics (ABS) national
census in 2021, are incorporated to include key social determinants such as housing conditions,
educational attainment, and employment. This data is rigorously overlaid with the 2021 hearing

48