Applying Analytics in an NPO

My name is John Huang, a current master student of business analytics (MSBA) in a school located in San Francisco. In the program, all the students will be split into groups to partner up with different organizations to apply analytics and provide values.

The students call the organization partners MIPs (MSBA Institution Partner). After the assignment process, I was assigned to a student team partnering it up with Angel Flight West (AFW) to utilize our analytical skills to create impact.

AFW is a non-profit organization which aims to provide free transportations for patients who cannot afford travel expenses to medical facilitations. The free transportations are all donated by pilots who want to fly as well as to help others. AFW serves as a platform to coordinate both pilots’ flights and patients’ requests. However, the matching processes are not very easy. It involves with many human labors to keep the processes smooth.

Knowing the fact, my team decided to provide our best efforts with analytics to alleviate some burdens in the current AFW’s system.

After several weeks, our team had gained some knowledge of AFW. We figured out, even many pilots were more than happy to help, a big portion of the pilots who signed up, were taking a long time to finish orientations or even a small portion only finished the orientation halfway and stopped. Because of the situation, AFW spent a lot of time and resources having staffs to reach out to those orientation-pending pilots to remind the pilots to finish orientation. The result was not significant.

Given the above situation, we decided to jump in to solve the discrepancy between invested resources and fruitless result to help AFW make frugal decision. We decided to first narrow our scope to only solve one of the fundamental problems, how long it would take for each pilot to finish his/her orientation processes by using machine learning.

We first extracted and cleaned the data from the pilots table of the database, and then bucketed the time span took for pilots to finish orientation. After manual bucketing the pilots, we started building our models based on relevant features. We tried decision tree, random forest and xgboosting models. At some point, we realized the proportion of the bucket was imbalanced. (imagine a credit card fraud situation which 99% credit cards are normal and 1% have fraudulent activities). The imbalanced data resulted the model to predict certain class fairly well while the prediction accuracy of other portions were not desirable. We decided to maneuver the imbalance by using Up-sampling technique to create balanced output. And now the model’s prediction accuracy has pumped a lot and is only required some tunings.

Once the model is fully developed and operationalize, AFW can save a lot of time and resources to focus on their goal. While acquiring new pilots are important, with the developed model, AFW can easily classify each new pilot into the bucket and allocate specific resource for that pilot to finish the orientation in a timely manner and serve the AFW’s goal — provide free transportations for patients.

This is our team’s first model and we have several more models under developments which can fully leverage the analytics power for AFW to make wiser decision to serve its goal.

Related posts:

About Fafafa Slots Hack Tool Fafafa slots is a game that has been recently released to IOS and Android devices all around the world. This game is a virtual slot gambling app, which features people…

Se acaba ya el 2018 y me ha apetecido volver a ver una de mis películas favoritas, Big Fish. Y que mejor manera que empezar a escribir en Medium que hablando sobre la obra magna de Tim Burton y…

2019 is upon us and now is usually the time of new year´s resolutions. So why not have a resolution that contributes to positive environmental actions? One way is to change the way you value and…