• 

  Education is the feature

📚 Behind the App: My Journey Building the Student Math Performance Predictor

🎯 What Inspired Me

As someone passionate about education and data science, I’ve always been fascinated by how background and environment influence student learning outcomes — especially in mathematics. I wanted to explore whether we could use simple, accessible indicators (like home resources and confidence levels) to predict a student's performance. This could help educators and parents better understand where support is needed most.

My goal was to build something that’s not only data-driven but also usable by non-technical users — like a teacher or parent — through an interactive web app.

🛠️ How I Built It

The project was developed using R and Shiny. Here's how I approached it:

🔍 1. Feature Selection

I began by training a regression model on a student dataset, but the R² score was only around 30%. That showed me that predicting exact performance scores wasn't reliable — so I pivoted to a classification model. To improve accuracy, I used Recursive Feature Elimination (RFE) to select the top 5 predictors that were most relevant to math performance:

• Student's confidence in math
• Home resources for learning
• Parents' education level
• Number of books at home
• Instructional hours per year

🧠 2. Model Training

I used caret with a Random Forest classifier, training it to predict High / Medium / Low performance categories instead of a numeric score. This turned out to be more robust and interpretable.

🌐 3. Web App with Shiny

I built a user-friendly Shiny app to allow users to:

• Input the values for each predictor using dropdowns and sliders
• Understand what each input means with clear descriptions
• Instantly see the predicted performance level

I also packaged the trained model as an .RData file so that the app could make predictions on the fly.

🧠 What I Learned

• Switching from regression to classification can make a big difference when prediction uncertainty is high
• Feature engineering and variable explanation are just as important as the model itself when building user-facing apps
• Building with Shiny helped me think about how to make machine learning more accessible
• Deploying with shinyapps.io taught me how to manage dependencies, versioning, and user

🔧 Challenges I Faced

• My first model had low predictive power — I had to rethink my approach
• Deploying to shinyapps.io was tricky due to package version conflicts and memory limits
• Some users didn’t understand the numeric input values, so I redesigned the UI with dropdowns and embedded descriptions for each variable
• Making the app lightweight enough to deploy without timeouts or crashes

✅ The Result

The final result is a responsive, educational app that predicts a student's math performance category based on five key inputs. It's built to be clear, user-friendly, and helpful for educators or researchers interested in exploring how background factors impact learning.

Built With

• r