🏎️ Dynamic Parking Pricing System (Real-time with Pathway & Bokeh)

This project implements a real-time dynamic parking pricing model that adjusts parking fees based on factors like demand, traffic, queue length, special events, and vehicle type. The goal is to optimize resource usage, reduce congestion, and improve user experience by simulating real-world parking dynamics using streaming data.

📍📍 Key Features

• ⏱️ Real-time processing using Pathway
• 📊 Interactive price visualizations using Bokeh + Panel
• 🧠 Two intelligent pricing models: Baseline and Demand-Based
• 📈 Prices stay fair: always between 0.5x and 2x of the base price

📂 Dataset Overview

The project uses a simulated real-time dataset that mimics the behavior of multiple city parking lots across different conditions.

🟥 Source File

• File Name: dataset.csv
• Format: CSV (Comma-Separated Values)
• Rows: 18,368
• Parking Lots: 14 unique locations (identified via SystemCodeNumber)

🔸 Schema

Column Name	Description	Type
SystemCodeNumber	Unique ID for each parking lot	string
Latitude	Geographic latitude of the parking lot location	float
Longitude	Geographic longitude of the parking lot location	float
LastUpdatedDate	The date when the occupancy data was last updated	string
LastUpdatedDate	The exact time on the update date when the data was recorded.	string
Occupancy	Current number of parked vehicles	int
Capacity	Total capacity of the parking lot	int
QueueLength	Number of vehicles waiting to park	int
TrafficConditionNearby	Nearby traffic status: Low, Average or High	string
IsSpecialDay	Indicates special events (e.g. holidays, festivals)	boolean
VehicleType	Type of vehicle: Bike, Car, Truck , Cycle	string

🔸 Engineered Features

These are derived during stream ingestion using Pathway and UDFs:

Feature	Description
OccupancyRate	Ratio of Occupancy / Capacity
TrafficLevel	Encoded traffic severity: Low → 0, Average → 1, High → 2
VehicleFactor	Encoded vehicle type: Cycle → 0, Bike → 1, Car → 2, Truck → 3
day_id	Unique ID per lot per day for time windowing
instance	Time-based grouping for real-time windowing
Timestamp	Event timestamp (format: %Y-%m-%d %H:%M:%S)

🛠️ Tech Stack

Component	Tool / Library	Role
Real-time Engine	Pathway	Stream processing & time windows
Visualization	Bokeh + Panel	HTML plots for pricing trends
Data Handling	Pandas, NumPy	Intermediate analytics
Storage	CSV, JSONL	Input stream and output data
Notebook	Google Colab	Dev & demo environment

🧠 Model 1: Real-Time Baseline Pricing Model

This model calculates a dynamic base price for each parking lot in real-time based on daily occupancy fluctuations. It uses Pathway's stream processing to aggregate data in daily tumbling windows.

⚙️ Key Idea

A parking lot’s demand volatility is captured using the difference between maximum and minimum occupancy rate within a day. Higher volatility suggests higher demand, which justifies price adjustments.

📈 Pricing Formula

 price = base_price + (occ_max - occ_min) / capacity 

🧱 Architecture Flow

1. Initial Ingestion: Load CSV data using Pandas
2. Preprocessing: Feature engineering including OccupancyRate and timestamp parsing
3. Streaming Pipeline: Stream preprocessed data via Pathway
4. Windowing: Daily tumbling window using windowby()
5. Aggregation:
   • Max/Min Occupancy Rate
   • Max Capacity
6. Pricing Calculation: Using the formula above
7. Export/Plotting: Saved to JSONL and visualized using Bokeh

📈 Results & Inference

• Goal: Adjust the parking price based on daily variations in occupancy.
• Observation: The plotted price fluctuates across days for each lot, depending on the difference between maximum and minimum occupancy rates.
• Inference: Lots with high variation in occupancy see greater pricing adjustments, reflecting dynamic demand. This model ensures basic responsiveness without overcomplicating the logic.

⚡ Model 2: Demand-Based Dynamic Pricing

This model adjusts parking prices based on real-world demand factors that affect parking pressure in urban areas. It provides a more responsive pricing scheme than the baseline.

🔍 Demand Formula

We define a composite demand function:

 Demand = α · (Occupancy / Capacity) + β · QueueLength − γ · TrafficLevel + δ · IsSpecialDay + ε · VehicleTypeWeight 

Where:

• α (alpha) = 5.0 → Emphasizes occupancy
• β (beta) = 0.4 → Importance of queue buildup
• γ (gamma) = 1.5 → Penalizes heavy traffic
• δ (delta) = 2.0 → Boosts price on special days
• ε (epsilon) = 1.0 → Based on type of vehicle (e.g., Truck > Car > Bike > Cycle)

💰 Price Formula

price = base_price × (1 + λ × NormalizedDemand)

• base_price = 10 (minimum baseline)
• λ (lambda) = 0.2 (scaling factor)
• NormalizedDemand = demand / max_demand, clipped to [0.0, 1.0]
• Price is clipped to remain within [0.5 × base, 2.0 × base]

🎯 Intuition

• Encourages higher prices when demand is high (e.g., long queues, special events)
• Offers lower prices when conditions are relaxed
• Ensures prices stay within a safe, bounded range

🔄 Architecture Flow

1. Inittial Ingestion: Load CSV data using Pandas.
2. Preprocessing:
   • Convert traffic and vehicle type to numerical levels.
   • Compute OccupancyRate and timestamp parsing.
3. Streaming pipeline: Stream preprocessed data via Pathway.
4. Windowing: Daily tumbling windows per lot
5. Aggregation:
   • Average Occupancy, Queue Length, Traffic, Vehicle Type
   • Special day indicator
6. Demand Calculation: Using the weighted formula
7. Price Calculation: Based on normalized demand
8. Export & Plot:
   • JSONL file written using Pathway
   • Visualized using Bokeh (per-lot pricing trends)

📈 Results & Inference

• Goal: Determine a more intelligent price based on a combination of factors — occupancy, queue length, traffic condition, vehicle type, and special day status.
• Observation: The plotted prices show more refined variation, often differing from Model 1, as they react to multi-factor demand changes.
• Inference: This model provides nuanced and fair pricing by factoring in local demand stressors. For example, high traffic and long queues during a festival day lead to higher prices.

---

🏗️ Architecture Diagram

---

🚀 Future Improvements

While the project currently implements two core dynamic pricing models — a Real-Time Baseline Model and a Demand-Based Pricing Model — a potential enhancement includes integrating a Competitive Pricing Model. This model would dynamically adjust parking rates based on nearby competitors’ pricing using geographic proximity (latitude and longitude). Additionally, future work could explore deeper integration of real-world traffic APIs, predictive occupancy trends, and user behavior analytics to further optimize pricing strategies and rerouting decisions.