title	emoji	colorFrom	colorTo	sdk	sdk_version	app_file	pinned
RideCastAI	🚖	purple	orange	streamlit	1.33.0	streamlit_app/app.py	false

🚖 RideCastAI — Smart Fare & ETA Estimation Engine for Ride-Hailing

🚀 RideCastAI is a smart, real-time fare and ETA prediction engine built using real NYC taxi data. It blends time-series forecasting, spatial demand modeling, and uncertainty-aware regression to simulate real-world ride-hailing behavior — deployable and production-ready.

---

🧠 What It Solves

"What will this ride cost me and how long will it take?"

This app uses real NYC taxi data and engineered ride context features like:

• Hour of day
• Day of week
• Trip distance
• Rush-hour indicators

to simulate dynamic pricing and duration estimation logic, similar to Uber/Lyft backend systems.

---

🔧 Features

✅ ETA prediction using XGBoost
✅ Fare estimation with uncertainty bands (Quantile Regression)
✅ NYC demand heatmap
✅ Streamlit UI with live trip simulation
✅ Model Prediction Latency printed in UI for performance observability
✅ Model performance metrics built-in
✅ Trained locally & deployed on Hugging Face Spaces
✅ Modular codebase for real-world use

---

🖼️ Architecture Overview

                        ┌────────────────────────────┐
                        │        User Input          | 
                        │ (Time, Day, Distance)      │
                        └────────────┬───────────────┘
                                     │
                                     ▼
                        ┌────────────────────────────┐
                        │    Feature Engineering     │
                        │ (rush hour, sin/cos, etc.) │
                        └────────────┬───────────────┘
                                     │
                                     ▼
                   ┌────────────────────────────────────────┐
                   │          ML Models (XGBoost, GBR)       │
                   │   ETA Model + Fare Models (Q10/Q50/Q90) │
                   └────────────┬───────────────┬────────────┘
                                │               │
                          ETA Prediction   Fare Prediction
                                │               │
                                ▼               ▼
                     ┌────────────────┐  ┌────────────────────┐
                     │   ETA Output   │  │ Fare Estimate (+/-)│
                     └────────────────┘  └────────────────────┘
                                │               │
                                └──────┬────────┘
                                       ▼
                          ┌───────────────────────────┐
                          │     Streamlit UI (App)     │
                          │ Predict Button triggers ML │
                          │ Heatmap visual via Plotly  │
                          └────────────┬───────────────┘
                                       │
                                       ▼
                   ┌──────────────────────────────────────┐
                   │   🔥 NYC Demand Heatmap              |
                   │       Real-time Plotly Density Map   │
                   └──────────────────────────────────────┘

---

🔍 Feature Engineering Highlights

• hour_sin, hour_cos: Circular encoding of time
• is_rush_hour: Encodes commute stress
• is_weekend: Affects ride supply/demand
• Quantile regression: Predicts fare range not just point estimate

✅ Makes the model explainable and closer to real-world behavior

---

📊 Evaluation

⚡ Prediction latency (in ms) is logged and displayed in the app UI on each request.

Model	Metric	Value
ETA Model	MAE	~3.9 mins
Fare Model	MAE (Q50)	~2.3 USD

Metrics auto-logged and displayed in app sidebar.

---

🧠 How It Works — Under the Hood of RideCastAI

RideCastAI isn't just a demo — it's a modular, production-grade ML platform designed to mirror real-world ride-hailing pricing engines. Here's how it works:

🚖 1. Feature Engineering from Raw Taxi Data

Raw trip data from NYC Yellow Taxi dataset is processed to extract meaningful features:

• Distance (in miles)
• Pickup hour (0-23)
• Day of week (0-6)
• Is weekend / Is rush hour
• Time cyclic encodings (sin/cos of hour)

✅ Why it matters: Features mimic how actual ride-hailing engines reason about traffic, demand, and surge pricing.

🧠 2. ETA & Fare Estimation Models

Two types of models are trained:

• ETA Model: XGBoost Regressor
• Fare Models: 3 separate GradientBoostingRegressor models for 10th, 50th, and 90th percentile fare estimates.

This helps simulate a fare range with uncertainty — like Uber's real-time quote system.

✅ Why it matters: Demonstrates quantile regression and uncertainty estimation for real-world predictions.

🧪 3. Streamlit App for Real-Time Inference

The frontend UI lets users input:

• Pickup hour
• Day of week
• Trip distance

And returns:

• ETA prediction
• Fare prediction (Low / Median / High)
• Prediction latency in ms
• Dynamic demand heatmap

✅ Why it matters: Showcases full-stack ML skill from model to UI deployment with latency logging.

---

🧪 Example Flow

1. User selects trip time and distance via Streamlit UI
2. App predicts:
   • ⏱️ ETA (in minutes)
   • 💸 Fare range (low–mid–high estimate)
3. 🔥 Demand heatmap of NYC shown for visual context
4. 📊 Sidebar displays live model performance (e.g., MAE)

---

📽️ Live Demo

👉 Try it on Hugging Face

---

🚀 Getting Started Locally

1. Clone the repo

git clone https://github.com/rajesh1804/RideCastAI.git
cd RideCastAI

2. Setup Python 3.10 (Recommended)

Use Python 3.10 for full compatibility.

3. Install Dependencies

pip install -r requirements.txt

4. Download Dataset and Preprocess

python src/download_dataset.py  
python src/preprocess.py

5. Train Models Locally

(venv) RideCastAI>python src\train_eta_model.py 
📥 Loading processed data...
🔀 Splitting into train/test...
🚀 Starting ETA model training...
✅ Training completed in 3.75 seconds   
📈 Evaluating model on test set...
📊 ETA MAE: 3.11 minutes
💾 ETA model saved to src/models/eta_model.pkl

(venv) RideCastAI>python src\train_fare_model.py
🔄 Loading and preprocessing data...
✅ Data loaded and split. Time taken: 5.58s

🚀 Training quantile model for q=0.1...
✅ Done. MAE @ quantile 0.1: 2.72 | Time taken: 441.53s
💾 Model saved to src/models/fare_model_q10.pkl

🚀 Training quantile model for q=0.5...
✅ Done. MAE @ quantile 0.5: 1.85 | Time taken: 483.59s
💾 Model saved to src/models/fare_model_q50.pkl

🚀 Training quantile model for q=0.9...
✅ Done. MAE @ quantile 0.9: 3.44 | Time taken: 442.07s
💾 Model saved to src/models/fare_model_q90.pkl

🎉 All quantile models trained and saved successfully.

5. Run Streamlit App

streamlit run streamlit_app/app.py

---

📁 Project Structure

RideCastAI/
├── assets/
│   └── ridecastai-architecture.png
├── data/
│   └── yellow_tripdata_2023-01.parquet
├── src/
│   ├── preprocess.py
│   ├── features.py
│   ├── train_eta_model.py
│   ├── train_fare_model.py
│   └── models/
│       └── *.pkl
├── streamlit_app/
│   ├── app.py
│   └── utils.py
├── requirements.txt
└── README.md

---

🧪 Tech Stack

• Frontend: Streamlit for interactive UI
• ML Models: XGBoost, GradientBoostingRegressor
• Metrics: Scikit-learn (MAE, metrics logging)
• Feature Engg.: NumPy, Pandas
• Deployment: Hugging Face Spaces

---

📍 Heatmap Note

The NYC demand heatmap is currently simulated using pseudo-random normalized ride volumes per zone. This can be replaced with real pickup zone distributions using TLC GeoJSONs.

---

📜 License

MIT — see LICENSE for details.

---

👋 Author

Built by Rajesh Marudhachalam
🏆 Uber, Lyft, Bolt, Grab
📧 Say hi at Gmail 🔗 LinkedIn