An end-to-end machine learning project that explores historical U.S. baby-name trends and turns the strongest findings into an interactive prediction website.

• Public demo: babynamesprediction.streamlit.app
• Current data coverage: 1880-2024
• Current deployment target: Streamlit Community Cloud on Python 3.10

This project asks a simple but engaging question:

Can historical naming patterns help predict whether a baby name is likely to become highly popular?

To answer that, I combined historical U.S. baby-name data with feature engineering, classification modeling, trend analysis, and a Streamlit app that lets users explore names and test prediction scenarios.

• Historical trend exploration using U.S. baby-name records
• Feature engineering based on name structure and popularity patterns
• Logistic regression modeling for Top 100 popularity prediction
• Additional clustering and time-series experimentation in notebooks
• A portfolio-ready Streamlit interface for trend exploration and prediction

This is more than a notebook-only capstone. It shows an end-to-end workflow:

1. Frame a question that non-technical audiences can understand
2. Clean and reshape a large historical dataset
3. Engineer features for modeling
4. Build a predictive workflow
5. Turn the analysis into a user-facing app

That combination makes it a strong data + product storytelling project.

The Streamlit app is organized into four sections:

• Overview: project framing, dataset size, and recent top-name examples
• Trend Explorer: compare names across time using count and ratio-based metrics
• Prediction Studio: generate a Top 100 prediction using the trained model
• Project Insights: summarize why the project is interesting and how it can improve

The app lives here:

• streamlit/app1.py

The main dataset used by the app includes:

• Name
• Year
• Gender
• Count
• Name_Ratio
• Gender_Name_Ratio

The current public app and packaged deployment dataset are updated through 2024.

The prediction workflow also uses engineered features derived in the modeling notebooks, including:

• Is_Famous
• Gender_Binary
• Rolling_Average_Gender_Ratio_5_Years
• Vowel_Count
• Ends_With_Specified_Letters

The current interactive prediction experience is centered on a logistic regression model trained to estimate whether a name is likely to land in the Top 100.

Current retrained model snapshot:

• Training window: 1880-2024
• Training rows: 145,000 top-1000 yearly records
• Test accuracy: 0.9785
• Test ROC AUC: 0.9953

Supporting experiments in the repository include:

• clustering analysis
• feature exploration
• time-series forecasting notebooks

For the portfolio version of the project, the app emphasizes the clearest value proposition:

trend exploration + popularity prediction

• data_source/ raw and supporting data files
• models/ trained model artifacts used by the app
• notebooks/ analysis, feature engineering, and modeling notebooks
• references/ research and background reading
• reports_slides/ course reports and presentations
• src/ supporting project notes
• streamlit/ interactive app code
• conda.yml project environment specification

• streamlit/app1.py
• reports_slides/An analysis of historical data and trends.pdf
• reports_slides/Machine Learning Models Analysis.pdf
• reports_slides/S3_Capstone_BabyName.pdf

This project includes both a Conda environment file and a pip-friendly requirements file:

• conda.yml
• requirements.txt

Typical local workflow:

conda env create -f conda.yml
conda activate capstone
pip install -r requirements.txt
streamlit run streamlit/app1.py

The easiest way to share this project publicly is through Streamlit Community Cloud.

Current production URL:

• https://babynamesprediction.streamlit.app/

Deployment settings:

• Repository root contains requirements.txt
• Entrypoint file: streamlit/app1.py
• Recommended Python version on Streamlit Cloud: 3.10

To make deployment practical on GitHub, the app reads a lightweight packaged dataset from:

• data_source/app_data.pkl.gz

If that file is not present, the app falls back to the larger local development dataset:

• notebooks/data.csv

If you want to refresh the app with the newest official SSA release:

1. Download the official SSA national data file (names.zip) or the yearly file (for example yob2024.txt).
2. Save the SSA birth totals page (numberUSbirths.html) locally from:
   • https://www.ssa.gov/OACT/babynames/numberUSbirths.html
3. Run:

python3 scripts/update_babyname_data.py \
  --source /path/to/names.zip \
  --birth-totals-html /path/to/numberUSbirths.html \
  --year 2024

This updates both:

• notebooks/data.csv
• data_source/app_data.pkl.gz

If you also want the prediction model to reflect the new year, retrain it after the data refresh:

python3 scripts/retrain_logistic_model.py

That rebuilds:

• notebooks/Featured_Data.csv
• models/preprocessor.pkl
• models/logistic_model.pkl

• Add screenshots or GIFs of the app for portfolio readers
• Make the famous-name feature source more transparent
• Link the app from a future portfolio website

Some larger datasets and model-related resources are also referenced through this Google Drive folder:

• Project data folder

Ying Zhou