College Recommendation System – Personalized Suggestions Using Neural Networks

Vision

This project was designed to help prospective students make informed college decisions by matching them with schools that align with their academic profile, personal preferences, and logistical constraints. The goal was to deliver a personalized, data-driven recommendation experience powered by neural networks and accessible via a web-based interface.

Approach

User Input Features:

• Desired Major

• College Area (rural to urban)

• Cost Range

• Distance from Home

• SAT Score & GPA

• Acceptance Rate Range

• Home Zip Code

Data Pipeline:

• Source: National Center for Education Statistics (NCES)

• Method: Used the requests library to fetch and compile institutional data.

• Processing:

  • Cleaned and normalized datasets using NumPy and Pandas

  • Handled missing values, encoded categorical variables

  • Conducted exploratory data analysis with Matplotlib

Neural Network:

• Architecture: Custom-built 2-layer neural network designed from scratch using core linear algebra and calculus concepts.

• Implementation: No external ML frameworks — relied solely on NumPy for matrix operations and training loop logic.

• Functionality: Ingests student profiles and outputs top-matching colleges based on multiple weighted criteria.

Full Stack Interface:

• Backend: Built with Flask to serve the model and handle user input.

• Frontend: Simple and clean UI for submitting preferences and viewing tailored college recommendations.

Challenges

• Building and tuning a neural network from scratch without ML libraries, while ensuring convergence and generalization.

• Handling large, messy datasets from NCES and adapting them to fit a model-ready format.

• Balancing user-defined weights across diverse metrics (e.g., location vs. cost vs. academic fit) in a way that felt personalized and fair.

Conclusion

The College Recommendation System effectively demonstrates the power of neural networks and clean data pipelines in building personalized, user-centric applications. By combining end-to-end machine learning development with full-stack deployment, the project showcases strong interdisciplinary skills in data science, software engineering, and user experience design.