Grade Predictor

🎓 AP Grade Predictor: Using Machine Learning to Forecast Student Performance

Project by: luojonah/palomarhealth_frontend
Related Issues: #32 - Show Saved Grade After Login, #28 - Grade Saves to Profile

---

🔍 Overview

Our AP Grade Predictor is a machine learning-powered web feature designed to give students an estimate of their grade based on their self-assessment. This project is integrated into the Palomar Health Frontend and leverages a trained regression model to generate both a percentage score and a letter grade. The results are saved to the user’s profile for easy retrieval.

---

🤖 The Machine Learning Model

We trained a Linear Regression model using a dataset of student characteristics and their actual grades. The dataset includes 11 input features, all scored from 1–5:

• Attendance
• Work Habits
• Behavior
• Timeliness
• Advocacy
• Tech Growth
• Tech Sense
• Tech Talk
• Communication and Collaboration
• Leadership
• Integrity

💡 Binarizing Input

Before feeding data into the model, we simplify (binarize) the input:

• Ratings of 1–3 become 0
• Ratings of 4–5 become 1

This gives a rough estimate of student behavior and performance trends without relying on overly fine-grained data.

🎯 Output

The model predicts a percentage score, which is then converted into a letter grade using standard grading thresholds:

90+ → A 80–89 → B 70–79 → C 60–69 → D <60 → F

---

🧠 Backend API

We built a REST API using Flask to serve the model:

POST /api/grade/predict GET /api/grade/predict (JWT-protected)

POST /predict

• Accepts JSON with an inputs list of 11 values (each between 1 and 5)
• Returns predicted percent and letter grade

GET /predict (JWT Protected)

• Returns the saved grade prediction from the user’s profile

This ensures that users can retrieve their grade later without needing to re-enter their inputs.

---

🔐 Authentication

We protect the GET endpoint using JWT-based authentication. Only logged-in users can retrieve their saved predictions, which are stored in the database under their user profile.

---

🧩 Frontend Integration

We tackled a few issues to make the frontend fully functional:

✅ Issue #28

We ensured that after a user submits their self-assessment, the predicted grade is saved to their profile.

✅ Issue #32

We added logic so that when users log back in, their previously saved prediction is displayed.

---

🌐 Tech Stack

• Frontend: HTML / Bootstrap (Palomar Health UI)
• Backend: Flask, Flask-RESTful, JWT Authorization
• ML: Scikit-learn (Linear Regression)
• Data: Custom CSV dataset of self-reported academic habits and grades

---

💻 Raw Backend Code (API + Model)

grade_api and Predict API Resource

from flask import Blueprint, request, jsonify, g
from flask_restful import Api, Resource
from api.jwt_authorize import token_required
from model.user import User
from model.grade_model import GradePredictionModel

grade_api = Blueprint('grade_api', __name__, url_prefix='/api/grade')
api = Api(grade_api)

model_instance = GradePredictionModel()

class Predict(Resource):
    def post(self):
        data = request.get_json()

        if not data or 'inputs' not in data:
            return {"error": "Missing 'inputs' field in JSON payload. Expected a list of 11 numbers (1-5)."}, 400

        user_input = data['inputs']

        if len(user_input) != 11:
            return {"error": f"Expected 11 input values, received {len(user_input)}."}, 400

        try:
            user_input = [int(val) for val in user_input]
        except ValueError:
            return {"error": "All input values must be numbers between 1 and 5."}, 400

        if not all(1 <= val <= 5 for val in user_input):
            return {"error": "Input values should be between 1 and 5."}, 400

        percent, letter = model_instance.predict(user_input)

        return jsonify({
            'predicted_percent': percent,
            'predicted_grade': letter
        })

    @token_required()
    def get(self):
        user = g.current_user
        return jsonify(user.grade_data)

api.add_resource(Predict, '/predict')

🔹 GradePredictionModel Class

import pandas as pd
from sklearn.linear_model import LinearRegression

class GradePredictionModel:
    def __init__(self):
        data = pd.read_csv("datasets/ap_predict_data.csv")
        data = data.dropna(subset=['Grade'])
        data.columns = data.columns.str.strip()

        grade_map = {'A': 90, 'B': 80, 'C': 70, 'D': 60, 'F': 50}
        data['GradePercent'] = data['Grade'].map(grade_map)

        self.features = ['Attendance','Work Habits','Behavior','Timeliness','Advocacy','Tech Growth',
                         'Tech Sense','Tech Talk','Communication and Collaboration','Leadership','Integrity']

        X = data[self.features]
        y = data['GradePercent']

        self.model = LinearRegression()
        self.model.fit(X, y)

    def predict(self, user_input):
        if len(user_input) != len(self.features):
            raise ValueError(f"Expected {len(self.features)} inputs, got {len(user_input)}")

        binarized = [0 if x <= 3 else 1 for x in user_input]
        percent = self.model.predict([binarized])[0]
        percent = max(0, min(100, percent))

        if percent >= 90:
            letter = 'A'
        elif percent >= 80:
            letter = 'B'
        elif percent >= 70:
            letter = 'C'
        elif percent >= 60:
            letter = 'D'
        else:
            letter = 'F'

        return round(percent, 2), letter