Project Proposal Feedback

[pooyanjamshidi]

Feedback on Healthcare Accessibility Proposal:

This is a compelling project, particularly in light of how transportation barriers can significantly limit healthcare access for vulnerable populations. Your approach of using Artificial Neural Networks (ANNs) to model complex relationships between transportation perceptions and healthcare accessibility is both timely and impactful. Below are my thoughts and suggestions to help refine and strengthen your proposal:

Strengths:

• Timely and Relevant Problem: Transportation as a barrier to healthcare is a pressing issue, especially for underserved populations. The focus on South Carolina is also crucial, as it grounds the project in a specific region with known disparities, making your project relevant and impactful at the state level.
• Data-Driven Approach: Leveraging data from the South Carolina Healthcare survey along with multidimensional built environment data from the EPA and socio-demographic data from the U.S. Census strengthens the project. Combining these datasets provides a comprehensive view of how transportation impacts healthcare accessibility.
• Innovative Use of AI: The use of ANNs to model these relationships is a strong choice, given their ability to capture non-linear and complex interactions between different variables. This could allow you to uncover deeper insights than traditional statistical models might miss.

Suggestions:

1. Clarify Data Preprocessing:

   • While you mention the use of datasets, it would be helpful to provide more detail on how you plan to preprocess the data before feeding it into the ANN. What kind of data cleaning, normalization, or feature engineering will be needed to ensure the model works effectively? Since your data comes from different sources (e.g., surveys, built environment data, census data), integrating and preparing this data will be a critical step.

2. Define ANN Architecture and Hyperparameters:

   • You’ve mentioned using TensorFlow for model development, which is a good choice. However, providing more specifics about the architecture of your ANN would strengthen your proposal. For instance, how many layers will your network have? What kind of activation functions will you use (ReLU, Sigmoid, etc.)? Will you apply regularization techniques like dropout to avoid overfitting? These details will give a clearer sense of how you plan to optimize your model.

3. Address Interpretability:

   • One potential challenge with ANNs is their interpretability. Since healthcare and transportation policy decisions rely heavily on clear, actionable insights, it may be beneficial to include techniques for interpreting your model’s outputs (e.g., SHAP values, LIME). This will allow you to better explain how certain transportation or demographic factors contribute to healthcare inaccessibility.

4. Model Evaluation and Metrics:

   • You mention using accuracy, precision, and recall, which are good starting points. However, given the context, it might also be useful to include metrics like F1 score or AUC-ROC, especially if your dataset is imbalanced (e.g., more people with good healthcare access vs. poor access). Additionally, cross-validation could be useful to ensure the robustness of your model.

5. Geospatial Analysis and Visualization:

   • One of the key deliverables you mention is detailed maps and visualizations. I would suggest being specific about what kind of geospatial analysis you’ll perform. Will you be using geographic information systems (GIS) to map transportation perceptions and healthcare access? How will you ensure that your visualizations are meaningful for policymakers? Consider using tools like QGIS or integrating your TensorFlow model with geospatial libraries in Python (e.g., GeoPandas).

6. Policy Implications:

   • This project could have real-world policy implications, so consider adding a section or a brief discussion on how the insights from your model could directly influence transportation or healthcare policies. For example, could your model be used to advocate for more accessible public transportation routes or targeted interventions in healthcare deserts?

Additional Thoughts:

• Model Comparison: While you are focusing on ANNs, consider comparing your ANN’s performance with simpler models like logistic regression or decision trees. This will allow you to demonstrate whether the complexity of the ANN is necessary for the problem at hand.

• Ethical Considerations: Since your project deals with sensitive demographic and healthcare data, it might be useful to explicitly mention how you’ll handle ethical concerns, particularly regarding privacy and bias. For instance, how will you ensure that your model doesn’t unintentionally reinforce existing inequalities?

• Scalability and Generalization: Once the model is developed for South Carolina, could it be generalized or applied to other regions? A discussion about how transferable your approach might be to other states or populations could be a valuable addition.

---

Conclusion:

Your project addresses a critical gap in healthcare accessibility through an innovative application of machine learning. By focusing on the relationships between transportation perceptions and healthcare access, you have the potential to generate meaningful insights that could influence both transportation and healthcare policy. My main advice would be to further specify the technical aspects of the ANN model, enhance interpretability, and think about the broader policy impact of your work.

Looking forward to seeing how this develops—great start!