Research Dashboard

Semantics-Aware Explainable Machine Learning for Urban Service Demand Forecasting in Smart Cities

An interactive dashboard for visualizing the SIMC NYC research paper. This tool helps partners, reviewers, and supervisors understand the spatiotemporal-semantic early-warning formulation for NYC urban service demand forecasting without reading the full paper first.

NYC 311 Data
Spatiotemporal Forecasting
Early-Warning System
SHAP Explainability
Method Pipeline Overview

Key Numbers

Scale of the NYC urban service demand forecasting dataset

30.9M+
Geocoded 311 Requests
262
Neighborhoods (NTAs)
575
Weekly Periods
9
Complaint Categories
9
Boroughs Covered

Research Questions

Three core questions driving this research

RQ1

Can a leakage-safe multi-source panel forecast next-week abnormal reported demand under a future chronological test?

RQ2

Do semantic service categories benefit from category-specific alert thresholds?

RQ3

Which temporal and urban-context feature groups drive predictions?

Key Contributions

What this research brings to urban service demand forecasting

Early-Warning Formulation

Spatiotemporal-semantic early-warning system for abnormal demand detection

Multi-Source Panel

NYC 311 + NOAA + OSM + PLUTO data integration

Chronological Evaluation

Target-week chronological evaluation with leakage-safe splits

SHAP Interpretation

Model explainability through SHAP-based feature importance analysis

Method Pipeline

Data sources → feature engineering → model training → evaluation → explainability

NYC 311 Requests
NOAA Weather
OSM POI
PLUTO Built Env
Feature Engineering
LightGBM / XGBoost
Ensemble + Thresholds
SHAP Explain

Model Performance Snapshot

Test set results for the selected ensemble model

0.3800
F1 Score
0.3012
Precision
0.5146
Recall
0.3294
PR-AUC
0.7661
ROC-AUC

Explore the Dashboard

Navigate to different sections for deeper analysis