agenthinkmesh/gcc_smartcity_ai

GCC Smart City & Infrastructure AI

Ten scikit-learn models for smart-city and infrastructure tasks framed around Kuwait: traffic flow, grid load, water-network leaks, waste collection, public-safety incidents, air quality, parking demand, infrastructure maintenance risk, emergency response time and building energy. One script, gcc_smartcity_ai.py, generates a synthetic dataset for each task (12,000 rows each). It then compares five candidate algorithms with 3-fold cross-validation and saves the best one per task as a joblib pickle. No real sensor, utility or incident data was used. The models are demonstrations and templates.

Author: AgenThink, Kuwait City

Models

Test metrics are on a 20% held-out split of the same synthetic data (random_state=42). CV is the mean 3-fold cross-validation score on the 80% training split. All numbers come from automl_results.json.

#	File	Task	Type	Selected algorithm	CV score	Test score	Test MAE
1	model_traffic_flow.pkl	Traffic Flow Prediction	Regression	Ridge	R² 0.8675	R² 0.8616	82.43 (volume)
2	model_grid_load.pkl	Smart Grid Load Forecasting	Regression	Ridge	R² 0.9630	R² 0.9613	91.17 (load units)
3	model_water_leak.pkl	Water Network Leak Detection	Classification	GradientBoosting	Acc 0.9663	Acc 0.9654	–
4	model_waste_collection.pkl	Waste Collection Optimization	Regression	Ridge	R² 0.9681	R² 0.968	4.08 kg
5	model_safety_incident.pkl	Public Safety Incident Prediction	Classification	GradientBoosting	Acc 0.9561	Acc 0.9546	–
6	model_air_quality.pkl	Air Quality Index Forecasting	Regression	Ridge	R² 0.8964	R² 0.8914	16.32 AQI
7	model_smart_parking.pkl	Smart Parking Demand Prediction	Regression	GradientBoosting	R² 0.9912	R² 0.9903	36.13 spaces
8	model_infra_maintenance.pkl	Infrastructure Maintenance Risk	Classification	GradientBoosting	Acc 0.9520	Acc 0.9525	–
9	model_emergency_response.pkl	Emergency Response Time Prediction	Regression	GradientBoosting	R² 0.9679	R² 0.9704	1.76 min
10	model_smart_building.pkl	Smart Building Energy Optimization	Regression	GradientBoosting	R² 0.9867	R² 0.9929	2.05 kWh

The candidates were RandomForest, GradientBoosting, ExtraTrees, DecisionTree, and either Ridge (for regression) or LogisticRegression (for classification). The CV score of every candidate is stored under all_scores in automl_results.json. The saved estimator classes match the best_algo values in the JSON.

Dashboard

The script generates this dashboard. It shows the test score for each model, plots of the synthetic data, feature importances and a heatmap of the algorithm comparison.

Repository files

File	Description
README.md	This model card
gcc_smartcity_ai.py	Full pipeline: synthetic data generation, AutoML selection, evaluation, dashboard, and saving of models and results
automl_results.json	For each model: task, type, selected algorithm, CV and test metrics, CV scores of all candidates, ordered feature list, feature importances, and a free-text gcc_note
gcc_smartcity_ai_dashboard.png	Performance and data dashboard (about 0.8 MB)
model_*.pkl (10 files)	Fitted scikit-learn estimators saved with joblib. The pickles record scikit-learn 1.8.0 as the training version

Training data

All training data is synthetic. The script generates it with NumPy (np.random.seed(42)); running it recreates the data. Each task has 12,000 independent rows and 14–16 features.

• Regression targets are hand-written formulas over a subset of the inputs, plus Gaussian noise, clipped to a range.
• Classification targets come from a hand-weighted risk score. A row is labelled 1 if its score is at or above a quantile: 0.82 for leaks (about 18% positives), 0.80 for safety incidents (about 20%) and 0.75 for maintenance risk (about 25%).
• Temperature inputs take only the discrete values {22, 28, 35, 40, 44, 46} °C.
• Several features are pure noise: they do not appear in the target formula. For example, expat_dense and economic_stress have no effect on the synthetic safety-incident label.
• Units for traffic volume and grid load are not stated in the code. The gcc_note strings in the JSON are the author's contextual remarks. They are not derived from data.

Features and targets

Important: these models were trained on plain NumPy arrays, so they store no feature names. Pass a 2-D array whose columns follow exactly this order, which matches features in automl_results.json.

1. Traffic Flow: target traffic volume (50–3,000). Inputs (16): hour, day_of_week, temp, rain, sandstorm, is_ramadan, is_eid, school_day, prayer_time, accident, road_work, pop_zone, signal_phase, merge_lanes, holiday, peak. peak is 1 when hour ∈ {8, 17, 18}, following the script's definition.
2. Grid Load: target load (500–8,000). Inputs (14): hour, temp, humidity, is_summer, is_ramadan, is_eid, population_k, prev_hour_load, solar_rad, ev_pct, industry_load, desalination_load, smart_meter_pct, ac_setpoint
3. Water Leak: target leak (0/1). Inputs (15): pressure, pipe_age, material, soil_type, temp, prev_leak, flow_anomaly, pressure_drop, night_flow, days_since_repair, corrosion_idx, pipe_depth, joint_type, vibration, salinity_soil
4. Waste Collection: target waste_kg (5–500). Inputs (14): zone, day_of_week, population, bin_fill_pct, temp, is_ramadan, is_eid, days_since_last_collect, commercial_pct, hospital_nearby, hotel_nearby, recycling_rate, smart_bin, food_waste_pct
5. Safety Incident: target incident (0/1). Inputs (15): hour, day_of_week, temp, is_ramadan, is_eid, crowd_density, outdoor_event, is_night, prev_incident, expat_dense, patrol_density, cctv_coverage, heat_stress, economic_stress, road_rage. Here is_night = hour < 4 or hour > 22, and heat_stress = temp > 42.
6. Air Quality: target AQI (20–500). Inputs (14): temp, humidity, wind_speed, sand_aod, sandstorm, traffic_vol, industry_emissions, oil_flaring, hour, month, inversion, desalination_em, construction_dust, humidity_inversion. Here humidity_inversion = humidity > 70.
7. Smart Parking: target occupied spaces (5 up to total_spaces). Inputs (14): hour, day_of_week, location_type, total_spaces, temp, is_ramadan, is_eid, nearby_event, is_peak, ev_charging, app_guidance, is_weekend, school_nearby, valet_available. Here is_peak = 10 < hour < 22. A code comment maps location_type 1–5 to mall, hospital, office, govt and mixed.
8. Infrastructure Maintenance: target high risk (0/1). Inputs (14): infra_type, age_years, condition_score, days_since_maintenance, traffic_load, salt_exposure, uv_exposure, sand_abrasion, heat_cycles_yr, budget_adequacy, prev_failures, design_standard, contractor_quality, subsidence_risk
9. Emergency Response: target response time in minutes (2–90). Inputs (14): incident_type, severity, hour, day_of_week, distance_km, traffic_level, sandstorm, is_ramadan, is_eid, units_available, road_quality, dispatch_time_min, gated_community, address_clarity
10. Smart Building: target energy_kwh, bounded between 0.02× and 0.25× floor area; the time period is not specified. Inputs (14): building_type, floor_area_sqm, occupancy_rate, outdoor_temp_c, solar_radiation, shading_coefficient, ac_cop, insulation_grade, smart_controls, has_solar_panels, occupancy_sensors, bms_installed, glazing_ratio, district_cooling

How to use

# pip install scikit-learn==1.8.* numpy joblib huggingface_hub
import json, joblib, numpy as np
from huggingface_hub import hf_hub_download

repo = "agenthinkmesh/gcc_smartcity_ai"
results = json.load(open(hf_hub_download(repo, "automl_results.json"), encoding="utf-8"))
model = joblib.load(hf_hub_download(repo, "model_traffic_flow.pkl"))

print(results["traffic_flow"]["features"])  # column order
hour = 17
x = np.array([[
    hour, 2, 44,          # hour, day_of_week, temp
    0, 1, 0, 0,           # rain, sandstorm, is_ramadan, is_eid
    1, 0, 0, 0,           # school_day, prayer_time, accident, road_work
    3, 90, 2, 0,          # pop_zone, signal_phase, merge_lanes, holiday
    int(hour in (8, 17, 18)),  # peak, as defined in the training script
]], dtype=float)
print("predicted traffic volume:", model.predict(x)[0])

The classifiers (water_leak, safety_incident, infra_maintenance) expose predict_proba. Class 1 is the high-risk / event class. Loading a pickle can execute arbitrary code, so only load files from sources you trust.

Re-running training

pip install numpy pandas matplotlib scikit-learn joblib
# The output directory is hard-coded: edit OUT="/home/claude/gcc_smartcity_ai" near the top of the script first
python gcc_smartcity_ai.py

The script regenerates the data, all 10 models, automl_results.json, the dashboard and a short README.md. It will overwrite a README in the output directory.

Intended use

• Examples and starting templates for tabular AutoML on municipal and utility problems with GCC-specific inputs (heat, sandstorms, Ramadan and Eid, prayer times).
• Teaching, prototyping and schema design before retraining on real, governed city data.

Out of scope: live traffic control, grid dispatch, emergency dispatch, public-health alerts or policing decisions.

Limitations and risks

• Synthetic data only. Scores show that the models recover the hand-written formulas in the generator. Real-world performance is unknown.
• Evaluation is in-distribution. The "forecasting" models were not tested on future periods or real sensor streams. Only accuracy is reported for classifiers, with no precision, recall or calibration, even though positive rates are about 18–25%.
• Public-safety model. Its inputs include expat_dense, an area indicator related to nationality, plus economic_stress and patrol density. Although expat_dense has no effect on the synthetic label, any real deployment of incident prediction for policing risks discriminatory outcomes and feedback loops. It requires legal review, bias auditing and human oversight, and it should not be used to target individuals or communities.
• Safety-critical domains. Emergency response, air-quality alerts and infrastructure risk affect public safety. Do not use these models in those settings without validation on real data.
• No feature names stored. Passing columns in the wrong order silently produces wrong predictions.

License

MIT

Citation

@misc{agenthink2026gccsmartcity,
  author       = {AgenThink},
  title        = {GCC Smart City \& Infrastructure AI},
  year         = {2026},
  publisher    = {Hugging Face},
  howpublished = {\url{https://huggingface.co/agenthinkmesh/gcc_smartcity_ai}}
}