rameshb79/mortgage-rate-predictor

Mortgage Rate ML - Pre-trained Models

This repository contains three pre-trained machine learning models for predicting and classifying mortgage rate movements.

Models Overview

1. Prophet Forecasting Model ⭐ PRIMARY

File: prophet_model.pkl (81 KB)

Highly accurate time-series forecasting model for predicting mortgage rates 4+ weeks in advance.

• Accuracy: MAE 0.1613%, RMSE 0.2024%, MAPE 0.0397%
• Best For: Exact rate forecasting, trend analysis
• Training Data: 1,044 weeks (2005-2024)
• Confidence: Very High (2.3x better than ARIMA)

Quick Start:

from predictor import MortgageRatePredictor

predictor = MortgageRatePredictor('prophet_model.pkl')
forecast = predictor.predict(periods=30)
print(forecast['dates'])      # 30 upcoming dates
print(forecast['predictions']) # Predicted rates
print(forecast['bounds'])      # Confidence intervals

→ Full Model Card

---

2. Classification Model

File: classifier_model.pkl (5 KB)

Random Forest classifier for binary UP/DOWN rate direction prediction.

• Accuracy: 100% (test set)
• Best For: Rate movement direction alerts, binary decisions
• Features: 9 engineered features (lags, moving averages, volatility)
• Note: Perfect accuracy indicates validation needed on real data

Quick Start:

import joblib
import numpy as np

classifier = joblib.load('classifier_model.pkl')
features = np.array([3.5, 3.4, 3.6, 3.7, 3.6, -0.1, -0.5, 3.55, 0.08])
prediction = classifier.predict(features.reshape(1, -1))
probs = classifier.predict_proba(features.reshape(1, -1))

print(f"Direction: {'UP' if prediction[0] == 1 else 'DOWN'}")
print(f"Confidence: {probs[0][prediction[0]]:.2%}")

→ Full Model Card

---

3. ARIMA Forecasting Model 🔄 FALLBACK

File: arima_model.pkl (10 KB)

Classical statistical ARIMA(1,1,2) model for backup forecasting.

• Accuracy: MAE 0.3611%, RMSE 0.4237%, MAPE 0.0905%
• Best For: Validation, sensitivity analysis, fallback
• Comparison: 2.3x less accurate than Prophet
• Advantage: Explainable, lightweight, no hyperparameter tuning

Quick Start:

import joblib

arima_model = joblib.load('arima_model.pkl')
forecast_result = arima_model.get_forecast(steps=4)
forecast_df = forecast_result.conf_int()
print(forecast_df)

→ Full Model Card

---

Quick Comparison

Feature	Prophet	Classifier	ARIMA
Accuracy	0.0397% MAPE ⭐	100% accuracy	0.0905% MAPE
Output	Exact values	UP/DOWN	Exact values
Speed	<1 sec	<100ms	<100ms
Confidence	Yes	Yes	Yes
Interpretable	Medium	High	High
File Size	81 KB	5 KB	10 KB
Best For	Primary	Binary alerts	Fallback

---

Installation & Setup

Option 1: Use Pre-trained Models

# Install dependencies
pip install pandas numpy scikit-learn statsmodels joblib

# Load models directly
import joblib
prophet = joblib.load('prophet_model.pkl')
classifier = joblib.load('classifier_model.pkl')
arima = joblib.load('arima_model.pkl')

Option 2: Use Predictor Wrapper

# Copy predictor.py from main project
cp ../src/predictor.py .

# Use the clean API
from predictor import MortgageRatePredictor
predictor = MortgageRatePredictor('prophet_model.pkl')
forecast = predictor.predict(periods=30)

---

Integration Guide

See INTEGRATION_GUIDE.md for complete instructions on integrating with MortgageRateNotifier.

---

License

MIT License - Free for commercial and personal use

---

Disclaimer

These models are provided as-is for educational and research purposes.

⚠️ Important Warnings:

1. Not for Financial Advice: Models should not be sole basis for financial decisions
2. Accuracy Unknown: Perfect classifier accuracy on test data likely indicates overfitting
3. Real-World Performance: Models trained on historical data; real-world performance uncertain
4. Requires Validation: Extensively test with real market data before production use
5. Fallback Plan: Always have ARIMA as fallback to Prophet

Use at your own risk. Validate thoroughly before deployment to critical systems.

---

Last Updated: April 8, 2026