Urban parking demand forecasting p6

Playground/Urban_Parking_P6/Use Cases/Urban_Parking_Demand_forecasting.json

@@ -0,0 +1,170 @@
+{
+  "use_case_id": "UC001",
+  "title": "Urban Parking Demand Forecasting Using Graph Neural Networks",
+  "author": "Aishwarya Ravichandran",
+  "team_members": [
+    "Chiranjeevi Veluri",
+    "Aditya",
+    "Aishwarya Ravichandran"
+  ],
+  "course": "Capstone Project",
+  "project_type": "Machine Learning and Graph Analytics",
+  "domain": "Intelligent Transport Systems (ITS)",
+  "description": "This use case presents a Graph Neural Network (GNN)-based solution for predicting next time-step parking occupancy in Melbourne. The system models parking bays as graph nodes and connects geographically nearby bays using edges. Temporal features such as hour of day and day of week are combined with spatial features including latitude and longitude to forecast whether each parking bay will be occupied or free. Users can select a Melbourne area and specify a desired time to generate customized predictions, which are displayed through an interactive graph visualization.",
+  "business_objective": "To develop an intelligent parking demand forecasting system that predicts future parking occupancy for selected Melbourne suburbs using Graph Neural Networks, helping drivers find parking more efficiently and supporting smart-city planning.",
+  "problem_statement": "Parking demand is influenced by both temporal patterns and spatial relationships between nearby parking bays. Traditional machine learning models treat each parking bay independently and fail to capture neighborhood effects. This project addresses the problem by modeling parking bays as a graph and predicting occupancy at the next time step.",
+  "dataset": {
+    "name": "Melbourne On-Street Parking Sensor Data",
+    "processed_file": "cleaned_parking_with_area.csv",
+    "source": "City of Melbourne Open Data Portal",
+    "attributes": [
+      "bay_id",
+      "timestamp",
+      "latitude",
+      "longitude",
+      "occupancy",
+      "area"
+    ],
+    "target_variable": "future_occupancy",
+    "areas": [
+      "Melbourne CBD",
+      "Richmond",
+      "Docklands",
+      "Carlton",
+      "Fitzroy",
+      "Brunswick",
+      "South Yarra",
+      "St Kilda"
+    ]
+  },
+  "features": {
+    "input_features": [
+      "hour",
+      "day",
+      "latitude",
+      "longitude"
+    ],
+    "temporal_features": [
+      "hour of day",
+      "day of week"
+    ],
+    "spatial_features": [
+      "latitude",
+      "longitude",
+      "graph connectivity"
+    ]
+  },
+  "forecasting_logic": {
+    "approach": "One-step forecasting",
+    "description": "Current features at time t are used to predict occupancy at the next time step t+1 using shift(-1)."
+  },
+  "graph_construction": {
+    "library": "NetworkX",
+    "nodes": "Unique parking bays",
+    "edges": "Parking bays within 0.5 km",
+    "distance_metric": "Haversine distance"
+  },
+  "model": {
+    "name": "Graph Convolutional Network (GCN)",
+    "framework": "PyTorch Geometric",
+    "hidden_units": 16,
+    "dropout": 0.3,
+    "output_classes": 2,
+    "optimizer": "Adam",
+    "learning_rate": 0.01,
+    "loss_function": "CrossEntropyLoss",
+    "epochs": 200,
+    "classification_threshold": 0.4,
+    "train_test_split": "80/20"
+  },
+  "performance": {
+    "train_accuracy": "~65%",
+    "test_accuracy": "~40-50%",
+    "evaluation_metrics": [
+      "Training Accuracy",
+      "Testing Accuracy",
+      "Prediction Probabilities",
+      "Predicted vs Actual Labels"
+    ]
+  },
+  "interactive_features": {
+    "area_selection": true,
+    "hour_input": true,
+    "day_input": true,
+    "clickable_visualization": true,
+    "node_details": [
+      "Bay ID",
+      "Predicted Status",
+      "Actual Status",
+      "Confidence Score",
+      "Latitude",
+      "Longitude"
+    ]
+  },
+  "visualization": {
+    "library": "Matplotlib",
+    "graph_library": "NetworkX",
+    "color_scheme": {
+      "red": "Actual occupied",
+      "green": "Actual free",
+      "yellow_border": "Incorrect prediction",
+      "blue": "Selected node"
+    }
+  },
+  "key_files": [
+    "graph.py",
+    "features.py",
+    "model.py",
+    "train.py",
+    "prepare_data.py",
+    "select_area.py",
+    "predict_by_area.py",
+    "visualize_predictions.py",
+    "cleaned_parking_with_area.csv",
+    "parking_gnn_model.pth"
+  ],
+  "technologies": [
+    "Python",
+    "Pandas",
+    "NumPy",
+    "NetworkX",
+    "PyTorch",
+    "PyTorch Geometric",
+    "Matplotlib",
+    "Scikit-learn"
+  ],
+  "limitations": [
+    "Moderate prediction accuracy",
+    "Limited feature set",
+    "One-step forecasting only",
+    "Static graph structure",
+    "No real-time data integration",
+    "Desktop-based interface"
+  ],
+  "future_improvements": [
+    "Integrate weather and event data",
+    "Use advanced GNN architectures such as GAT",
+    "Support multi-step forecasting",
+    "Hyperparameter optimization",
+    "Real-time API integration",
+    "Web dashboard deployment",
+    "Parking recommendation system"
+  ],
+  "repository_structure": {
+    "src": "Python source code",
+    "data": "Processed datasets",
+    "models": "Saved trained models",
+    "docs": "Use case documentation"
+  },
+  "tags": [
+    "Graph Neural Networks",
+    "Urban Parking",
+    "Machine Learning",
+    "Forecasting",
+    "Smart Cities",
+    "Intelligent Transport Systems",
+    "PyTorch Geometric"
+  ],
+  "difficulty": "Intermediate",
+  "status": "Completed"
+}