PRAVAH

Predictive Adaptive Traffic Signal System

A Python prototype that simulates intelligent traffic signal control using machine learning — demonstrating measurable improvement over fixed-timer signal systems used in Indian cities today.

Built as part of a smart infrastructure research initiative aligned with NHAI's traffic management objectives.

---

Result

PRAVAH reduced average vehicle wait time by 14.4% during peak hours and 5.1% overall in a simulated urban intersection environment.

Peak hours (7–9am, 5–7pm) show the highest gains — exactly when congestion costs the most. Off-peak hours intentionally favour simpler logic, since adaptive overhead isn't justified at low volumes. This is a deliberate design choice, not a limitation.

---

What It Does

Component	What it solves
Data Generator	Simulates 30 days of realistic 24hr traffic (2,880 records) with morning/evening peaks
Prediction Model	Polynomial regression with cyclical time encoding — predicts vehicle count per 15-min interval
Signal Logic	Compares fixed 30s green (current standard) vs PRAVAH adaptive green (scales with predicted load)
Visualizer	3-panel dashboard: traffic volume, signal timing, wait time comparison

---

System Architecture

traffic_data.csv
      │
      ▼
data_generator.py     ← synthetic 24hr traffic with realistic peaks
      │
      ▼
model.py              ← polynomial regression, cyclical feature encoding
      │  (predicted volume)
      ▼
signal_logic.py       ← adaptive_signal() vs fixed_signal() + wait time sim
      │  (simulation results)
      ▼
visualize.py          ← 3-panel matplotlib dashboard

---

How to Run

Requirements

pip install pandas numpy scikit-learn matplotlib

Run full pipeline

python main.py

This runs all 4 phases in sequence and generates all output files.

Run individual modules

python data_generator.py   # generate traffic data only
python model.py            # train and evaluate model only
python signal_logic.py     # run simulation only
python visualize.py        # generate plots only

---

Output Files

File	Description
traffic_data.csv	2,880-record synthetic dataset
model.pkl	Trained regression model
simulation_results.csv	Hour-by-hour fixed vs adaptive comparison
pravah_results.png	Full visualization dashboard

---

Model Details

• Algorithm: Polynomial Regression (degree 4) via scikit-learn Pipeline
• Features: time_of_day, hour_sin, hour_cos (cyclical encoding), is_peak
• Why cyclical encoding: Hour 23 and hour 0 are numerically distant but traffic-wise similar. sin/cos encoding preserves this continuity.
• Performance: MAE ≈ 7.3 vehicles | R² ≈ 0.937

---

Signal Logic

# Fixed (current reality)
green = 30s  # always, regardless of traffic

# PRAVAH Adaptive
ratio = predicted_volume / max_expected_volume
green = MIN_GREEN + ratio × (MAX_GREEN - MIN_GREEN)
green = clamp(green, 15s, 75s)

Wait time is estimated using a Webster's formula approximation — standard in traffic engineering literature.

---

Project Structure

pravah/
├── main.py                  ← runs full pipeline
├── data_generator.py        ← synthetic traffic data
├── model.py                 ← prediction model
├── signal_logic.py          ← adaptive vs fixed simulation
├── visualize.py             ← dashboard plots
├── traffic_data.csv         ← generated dataset
├── simulation_results.csv   ← comparison output
├── pravah_results.png       ← visualization
└── README.md

---

Limitations (honest)

• Single intersection simulation — real networks have cascading effects
• Synthetic data — real deployment needs live sensor/camera feeds
• No vehicle classification — trucks and motorcycles have different discharge rates
• Webster's approximation — full microsimulation (SUMO, VISSIM) would be more precise

These are known next steps, not unknown gaps.

---

Alignment with NHAI Objectives

NHAI Priority	PRAVAH Contribution
Reduce urban congestion	Adaptive signal reduces peak wait time by 14.4%
Data-driven infrastructure	ML model trained on traffic patterns
Emergency vehicle priority	Architecture supports green corridor extension
Scalable deployment	Modular design — each component is independently replaceable

---

Developed by Aryan — Delhi Technological University