This project, completed by Jay Gleeson and Victoria Uriostegui was performed with the help of the CIC | PCUBED organization at Cal State Fullerton.
According to research completed by the Department of Education, while high school graduation rates have been increasing on average in the United States, lower-income households’ graduation rates are declining.¹ Thus, the main goal of the research was to see if there was a significant bias based on that statistic. The hypothesis developed was whether graduation rates correlate with household income statistics and if they could be predicted using various data prediction models.
Various tools were utilized for the project. After initially using Anaconda as the main virtual environment, Google Colab was eventually settled on, a cloud-based tool that matched the use-case. Within Google Colab, a Jupyter Notebook environment was set up, loaded with NumPy, Pandas, and TensorFlow packages. Also utilized was Seaborn, for making a heatmap to determine correlation rates between data columns. NumPy provided manipulable arrays and tensors, Pandas provided tools to clean data effectively, and TensorFlow simplified means to creating a neural network. The three models settled on were a neural network, a Decision Tree Regressor, and a Random Forest Regressor. Scikit-learn quickly became an invaluable tool for splitting the dataset into training and prediction data, as well as providing the library for the Decision Tree and Random Forest models. Finally, Plotly, was used to create an interactive US map, and Matplotlib, was used to create various data visualization graphs.
Note the database utilized in this Research Project was acquired through Kaggle and was created by Golden Oak Research Group in Orlando, Florida.²
Most recently, the previous logistic regression models were all swapped out with regression models to fit more accurately with the database's high school graduation data. Thus, after developing the models, training, and confirming predictions and accuracy scores, what was ended up with was very promising low error rates for each of the models. The neural network ended up having a Mean Absolute Error (MAE) rate of 0.0718, the Random Forest Regressor had an MAE rate of 0.0682, and the Decision Tree Regressor had an MAE rate of 0.0678. R Squared (R2), Root Mean Squared Error (RMSE), and Mean Percent Error (MPE) rates can be viewed within the .ipynb. Thus, given low error rates and sufficient model accuracy, it can be firmly stated that is there is indeed a predicable correlation between household income factors and high school graduation rates.
- Clone the repo
git clone https://github.com/jay-gleeson/graduation-prediction-csuf.git
cd graduation-prediction-csuf- Setup virtual environment.
python -m venv venv
venv\Scripts\activate
pip install -r requirements.txt- Launch Jupyter Notebook.
jupyter notebook- Open .ipynb in Google Colab. See: Open in Colab.
- Run Jupyter Notebook via Runtime >> Run All or Ctrl+F9.
Below is a copy of the original presentation for the Project RAISE Summer Research Symposium 2024 at California State University Fullerton.
¹ National Center for Education Statistics, “High School Graduation Rates.” U.S. Department of Education, Institute of Education Sciences, 2024, https://nces.ed.gov/programs/coe/indicator/coi.
² Golden Oak Research Group. “Insightful & Vast USA Statistics.” Kaggle.com, Alexander Geiger, 2018, https://www.kaggle.com/datasets/goldenoakresearch/us-acs-mortgage-equity-loans-rent-statistics.