AI Passenger Flow Forecasting for Transit

Picture a metro dispatcher seeing on the dashboard that in one hour, the 'Sportivnaya' station is expected to have an 85% surge due to a concert. They have 45 minutes to reroute an extra train and boost staff. This is reality with our AI system for passenger flow forecasting. We have developed and implemented such solutions for bus depots, subways, and railway operators. Forecast accuracy directly impacts operational costs: reduced downtime, optimal scheduling, and decreased overtime. Our models achieve MAPE 8-12% for a 1-hour horizon — 2-3 times more accurate than classical statistical methods. We take into account seasonality, weather, city events, and network topology. For example, on one station we reduced overtime by 30%, saving about 2 million rubles per year. The system pays for itself in an average of 6 months, delivering net savings from 1.5 million rubles annually.

Main Forecasting Tasks

• Optimize headways: when a peak is expected, the system recommends reducing the interval from 3 to 1.5 minutes.
• Schedule staff: flow forecasts per station allow calculating the number of cashiers and inspectors per shift.
• Prevent overcrowding: early warning 60-90 minutes before an anomaly.

How We Build Models

We use a combination of gradient boosting (LightGBM) and graph neural networks (GNN) for metro. For buses and railways, LightGBM with rich feature engineering is often sufficient.

# LightGBM with rich feature set
features = {
    # Lags
    'passengers_lag_15min': passengers_t_minus_1,
    'passengers_lag_1h': passengers_t_minus_4,
    'passengers_same_time_yesterday': passengers_same_period_yesterday,
    'passengers_same_time_last_week': passengers_same_period_week_ago,

    # Time
    'hour': hour,
    'minute': minute,
    'day_of_week': dow,
    'is_holiday': holiday_flag,
    'month': month,

    # External
    'weather_rain': rain_intensity,
    'temperature_c': temperature,
    'stadium_event_distance_time': event_proximity_score,

    # Station/route
    'station_type': encode(terminal_transfer_intermediate),
    'line_id': line_embedding
}

Why Graph Neural Networks Are Effective for Metro

In metro, flow at one station strongly depends on neighboring stations — passengers transfer, a line closure redistributes load. GNN models this dependency explicitly, yielding a 2-3% MAPE gain at peak hours. LightGBM processes data 10x faster than GNN during training, but GNN consistently wins on tasks with explicit network structure. Certified engineers with over 5 years of experience tune hyperparameters for each project.

Compare accuracy of different approaches on one dataset:

What Does Event Awareness Provide?

Without event-awareness, the forecast on a concert day deviates by 30-50%. We add known future covariates: for planned events — a flag and duration, for anomalous ones — automatic detection via z-score. This reduces peak-hour error to 12%. Thanks to our AI forecasting system for passenger flow, we consistently achieve high accuracy even in non-standard situations. Our AI passenger flow forecasting system includes an event detection module based on LightGBM and GNN.

Typical Implementation Mistakes

• Ignoring events: without an event calendar, peak forecast error is 30-50%.
• Missing lags: flow strongly correlates with previous intervals.
• Overly complex model for small data: GNN requires >6 months of history.

Scope of Work

• Data audit: analyze AFC, CCTV, IoT sources and build ETL pipeline.
• Baseline model: LightGBM in 2 weeks with MAPE 10-14%.
• Model enhancement: if needed, GNN or TFT, training on GPU cluster.
• Integration: control center dashboard with heatmap, forecast API, alerts on threshold exceedance.
• Testing: A/B test on a station, compare with current methods.
• Documentation and training: feature list, usage manual, model handover.
• Support: monitor data drift, retrain monthly.

Implementation Phases: From ETL to Dashboard

1. Audit and ETL: analyze sources (AFC, CCTV, IoT) and build processing pipeline.
2. Baseline model: LightGBM in 2 weeks, accuracy MAPE 10-14%.
3. Enhancement: if needed, GNN or TFT, training on GPU cluster.
4. Integration: control center dashboard with heatmap, forecast API, threshold alerts.
5. Testing: A/B test on a station, compare with current methods.
6. Deployment and support: monitor data drift, retrain monthly.

Estimated Timelines

• Pilot on one station/route — 3-4 weeks.
• Full network system with GNN and dashboard — 4-5 months.
• Costs are calculated individually. Request a pilot project on one station to evaluate the effect. Get a consultation: we will help you choose the optimal solution for your transport.

When does a time series forecasting model fail in production?

The CFO requests a quarterly sales forecast. An analyst builds SARIMA on three years of data, achieves MAPE 8.3% on the test set, and deploys. Two months later, the metric in production jumps to 23%. The root cause: the model was trained on pre‑COVID data, tested on a stable period, but production hit a promotion and supply chain disruption. Data leakage plus distribution shift—perfect notebook numbers, a broken forecast in reality. We have seen this pattern dozens of times across retail, fintech, and IoT. Our team has delivered more than 50 forecasting projects over 5+ years.

Incorrect cross-validation. Standard train_test_split for time series creates data leakage: the model sees future values during training. The correct approach is TimeSeriesSplit or walk‑forward validation with an expanding window.

Multiple seasonality. Hourly electricity consumption has three seasonalities: daily (24h), weekly (168h), yearly (8760h). SARIMA handles only one. Prophet can handle multiple but scales poorly to thousands of series.

Missing values and anomalies. A missing sensor reading is information (the sensor turned off), not NaN. Linear interpolation destroys this signal. Proper handling depends on the missingness mechanism.

Cold start. A new SKU in a 50,000‑item assortment has no history, yet a forecast is needed. Standard approaches fail; cross‑learning or feature‑based methods are required.

Why is model selection critical for your data?

Prophet (Meta) – a solid start for business data with clear seasonality and holidays. Fast setup, interpretable, built‑in outlier detection. Fails on irregular patterns and does not scale beyond ~10k series without parallelization.

Gradient boosting on features (LightGBM, XGBoost) – often underestimated. Engineer lags (t‑1, t‑7, t‑28), rolling means, day‑of‑week, holidays. The model trains on all series simultaneously, solving cold start via transfer learning. MAPE in retail often beats neural nets with proper feature engineering.

TFT (Temporal Fusion Transformer) – a transformer designed for interpretable forecasting with covariates. Built‑in variable selection, temporal attention, quantile outputs. Available in pytorch‑forecasting. Requires ~10,000+ records per series for stable training.

PatchTST – splits the series into patches (like ViT for images), capturing local patterns better than classic transformers. Excellent for long‑horizon forecasting (96–720 steps ahead).

N‑HiTS, N‑BEATS – attention‑free neural architectures, faster than TFT, competitive accuracy. N‑BEATS won the M4/M5 benchmarks for tasks without covariates.

How do we deploy TFT in production?

A typical pipeline via pytorch‑forecasting:

training = TimeSeriesDataSet(
    data,
    time_idx="time_idx",
    target="sales",
    group_ids=["store", "sku"],
    min_encoder_length=max_encoder_length // 2,
    max_encoder_length=max_encoder_length,  # 120 days
    min_prediction_length=1,
    max_prediction_length=max_prediction_length,  # 28 days
    static_categoricals=["store_type", "category"],
    time_varying_known_reals=["price", "promo_flag"],
    time_varying_unknown_reals=["sales"],
    target_normalizer=GroupNormalizer(groups=["store", "sku"], transformation="softplus"),
)

A common mistake: the default target_normalizer (StandardScaler) breaks predictions for series with zero values (no sales on weekends). GroupNormalizer with transformation="softplus" is the correct choice for count data.

Case study: retail demand forecasting

A chain of 120 stores, 8,000 SKUs, 28‑day forecast horizon. The original system: SARIMA per series, MAPE 18.4%, retraining cycle – 6 hours. We replaced it with TFT on PyTorch + pytorch‑forecasting: a single model for all series, MAPE 11.2%, retraining – 40 minutes on an A10G. Feature importance via variable selection revealed that day_before_holiday influences more than the holiday date itself. Annual savings on inference alone exceeded $50,000.

Step‑by‑step configuration

1. Data collection and preparation. Handle missing values (mark NaN, interpolate only for technical failures), aggregate to required frequency, engineer covariates (holidays, promotions, prices).
2. Create TimeSeriesDataSet. Set group_ids (store + SKU), time index, forecast horizon. Choose target_normalizer based on target distribution.
3. Train a baseline. Prophet or LightGBM first – to understand complexity.
4. Train TFT. Use TemporalFusionTransformer with loss=QuantileLoss(), tune learning rate and hidden layer sizes.
5. Validate and interpret. Walk‑forward test, analyze variable selection, build attention heatmaps.

How to properly evaluate forecast quality?

RMSE alone is misleading – it over‑penalizes large values. Our standard set:

• MAPE – interpretable, unstable near zero.
• sMAPE – symmetric, avoids division by small numbers.
• MASE (Mean Absolute Scaled Error) – normalized relative to a naive seasonal forecast, ideal for comparing series of different scales.
• Pinball loss – for probabilistic forecasting, inventory management.

We guarantee a model card with these metrics on the validation set and walk‑forward results on at least 6 months of history.

What deliverables do you receive?

• Documentation of chosen architecture and hyperparameter rationale.
• Reproducible training and inference pipeline (Docker + CI/CD + Airflow/Prefect).
• Committed code with unit tests for key components.
• Team training: retraining, output interpretation, deployment of new versions.
• 3 months of post‑delivery support (consultations, bug fixes, fine‑tuning).

The model is deployed via FastAPI or Triton Inference Server. Retraining is scheduled (e.g., weekly) via Airflow with drift validation and automatic rollback if metrics deteriorate.

Process and timeline

We start with EDA: visualization, ADF test, STL decomposition, analysis of missing values and outliers. This takes 2–3 days but often reveals systemic data issues that block forecasting. Then we build a baseline (naive seasonal, Prophet), engineer features for LightGBM, and select a neural architecture if needed. Walk‑forward validation with a realistic horizon. Deployment via API with automatic retraining scheduled via Airflow or Prefect.

Timeline: MVP forecast on one data type – 3–6 weeks. Hierarchical forecasting system with automation – 2–5 months. Cost is calculated individually based on data volume, number of series, and required accuracy.

Our team consists of certified ML engineers (AWS ML Specialty, GCP Professional ML Engineer) with 5+ years on the market and over 50 completed forecasting projects. Contact us for a free analysis of your data – we will assess the task and provide initial recommendations within 1–2 days. Request a consultation to ensure your forecasts work in production, not just in a notebook.