Inspiration
First of all, the vision of Medlink to provide low-cost health care to the people of Charlotte is very inspiring. But it is very depressing that there is a gap in access to these services. We, as a team of aspiring data scientists, looked into the problem and brainstormed for a data-centric approach which can be able to find the prime factors which are affecting the people from not availing these services provided by Medlink.
What it does
Our project identifies the economic factors affecting the people in various areas of Charlotte. It makes an area-wise(using zip code) analysis of these economic factors and predicts the prime factors among those factors which made the people not use the services provided by Medlink. It can be helpful for the organization to carefully look for these factors in an area and thus can create better awareness to the people in those areas about Medlink's services.
We developed two predictive models which predict the number of people in a particular area who potentially should be using the Medlink services in future when provided the economic factors in that area. These models serve the system by predicting the actual number of patients who could be visited every week/month that can be crosschecked with the actual patients visited that week/month to fill up the gap.
We considered some important economic factors that we assumed to be crucial for the lives of people. But, we developed our model in such a way that it can be scaled up to analyze other economic or social or any potential factors that can be causing this problem.
How we built it
We focussed more on extracting the crucial data from multiple sources of data provided. We then processed the data and prepared our dataset, which conveys the economic factors affecting people in particular areas of Charlotte and also the number of people using the services of Medlink from those areas.
For our first model, we used Random Forest Regressor to predict the potential number of people that can be visiting the health care centers. This model also explains the most critical features that are affecting the number of people using health care services. For our second model, we developed a predictive model, leveraging our knowledge of deep learning, to predict the same outcome as our previous model.
We used Python as our primary coding language with libraries like numpy, pandas and sklearn. Also, we designed our neural network with Tensorflow library. We used Tableau, matplotlib for visualizing our analysis.
Challenges we ran into
The very first challenge we ran into is forming the question that the data can answer and that can potentially solve the Medlink's problem. Once, we figured it out, our main challenge was to prepare a dataset from a wide range of available data sources. The preprocessing was very tedious as we decided to approach based on the zip codes in Charlotte, so mapping the locations and zip codes took time.
After obtaining the clean data, it would have been great if we could get recent census data, which could give us a robust trained model ready to deploy. We are also restricted with the less number of rows in the data as we are only looking at Charlotte city, and when expanded our model can perform even better.
Accomplishments that we're proud of
We are proud that we developed a model that can predict the patient's visiting pattern based on their area. We are also very content with the dataset we prepared that can be utilized in the future.
In the end, we feel accomplished that we are awake all night and completed what we planned to.
What we learned
We learned the importance of data analytics in Health care and how data can change the way health care operates to provide better health to people.
We learned that we should not give up even if the internet stops for some time.
What's next for WitWorkers
We look forward to collecting more data from various cities in USA and develop a robust model.
Log in or sign up for Devpost to join the conversation.