AI System for Hospital Occupancy Forecasting

In a 200-bed hospital, chaos every Monday. Our AI solution predicts load hourly, enabling proactive resource reallocation, cutting waiting time by 30% and saving over $200,000 annually through staff optimization. Manual planning yields 30% MAPE; our system delivers under 10% — three times better. Accurate occupancy prediction is key to profitability.

Our AI in healthcare approach combines machine learning hospital models for occupancy, LOS prediction, and hospital resource planning, all visualized in a medical dashboard with MIS integration — achieving healthcare cost reduction and MAPE accuracy. Without an accurate forecast, you cannot efficiently allocate beds, plan discharges, or avoid idle expensive equipment.

Data for Occupancy Prediction

Our models rely on historical admission series, weather, epidemiological situation, and calendar factors. Key groups:

• Incoming flow: admissions to ER (by hour), planned hospitalizations by profile, discharges, length of stay (LOS)
• Resources: staffing needs by department, OR load, consumables (ventilators, medications), lab tests

Weather data (temperature, humidity, pressure) are significant predictors for cardiology and pulmonology. Epidemiological indices (e.g., Flu Index) are published with 1-2 week delay, so we use proxies—search query volumes.

Why an Ensemble?

For regular patterns with yearly and weekly seasonality, we use SARIMA. ML models—LightGBM with lagged features, weather, and epidemiological predictors—improve accuracy by 20-30% (up to 1.3 times better):

We avoid look-ahead bias: during training with time shift, we account for the delay in epidemiological data. Ensemble models yield 5% higher accuracy than single models—confirmed over 15 projects.

How to Improve the Accuracy of Length of Stay Prediction?

For LOS (Length of Stay) we use survival analysis with covariates: ICD-10 diagnosis, age, sex, Charlson Comorbidity Index, admission type, initial lab results. We apply Accelerated Failure Time (AFT). Accuracy: MAE 1.5-2.5 days at average LOS 5-7 days. This enables better bed turnover planning: predicting LOS one day more accurately gives +3% bed utilization efficiency.

Resource Planning Based on Forecast

1. Staff: Nurses_needed = ceil(Expected_Patients / Nurse_Patient_Ratio). Ratio depends on department (ICU 1:2, general ward 1:8).
2. Operating Rooms: forecast of elective and emergency surgeries + CP-SAT optimization of schedule considering teams, equipment, and duration.
3. Supplies: forecast consumables via regression on expected patient volume and procedure types. We integrate with pharmacy systems (1С:Больничная аптека) for automatic reorder at ROP.

Dashboard for Hospital Management

Operational screen for the chief physician: current load vs. forecast by department, alert for bed/staff shortage risk at 24/48/72 hours. Strategic dashboard for administration: accuracy metrics, seasonal patterns, capacity expansion analysis. Integration with MIS via HL7 FHIR (ЕМИАС, SAMSON, MedElement). We follow MLOps hospital practices to maintain model performance.

What's Included

• Hospitalization forecasting module (MAPE <10% after stabilization)
• Length of stay prediction model (MAE 1.5-2.5 days)
• OR schedule optimizer and staff allocation tool
• Dashboards for chief physician and administration
• MIS integration via HL7 FHIR / REST API
• Technical documentation and staff training (2 days onsite)
• 6 months post-launch support

Implementation Process

1. Data audit, cleaning, and preparation — 2-3 weeks.
2. Base model development and training (≥12 months history) — 6-8 weeks.
3. Full system build: LOS prediction, OR optimization, dashboards — 4-5 months.
4. Testing and calibration for the specific department — 2 weeks.
5. Handover, staff training, documentation — 1 month.

We guarantee forecast MAPE <10% after stabilization (typically 3 months). Average budget savings up to 15%, which for a 200-bed hospital translates to over $200,000 annually. Get an AI system for your hospital—contact us to discuss your case.

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.