AI Health Monitoring for Livestock: From Sensors to Predictions

On a farm with 200 heads of cattle, over 50,000 records are generated daily from ear tags, boluses, and milking robots. Raw data is just numbers. Without ML processing, you risk missing the onset of mastitis 48 hours before symptoms or not detecting estrus, costing tens of thousands of rubles per lost pregnancy. Our AI system turns these numbers into precise alerts and forecasts. We specialize in integrating with existing sensors and customizing models for your herd. By early mastitis detection, savings reach 1.5 million rubles per 100 heads per year. In this article, we'll break down key problems and show how ML solves them more effectively than traditional methods.

What problems does AI monitoring solve?

Estrus (heat) – timely detection is critical for reproduction. Threshold methods using a single accelerometer yield 70–80% accuracy. ML on multi-sensor data (activity + rumination + temperature) raises detection rate to 95%, reducing misses.

Mastitis – udder inflammation: drop in rumination, decreased milk yield, and increased temperature. The ML model detects it 24–48 hours before clinical symptoms using a combination of signs from bolus and milking robot. This allows early treatment and reduces milk loss.

Subacute ruminal acidosis (SARA) – direct monitoring of rumen pH via bolus. If pH < 5.8 for more than 3 hours a day – alert. ML predicts risk based on feeding data, giving time for ration adjustment.

Lameness – reduced activity, asymmetric steps, slow walking. ML model assesses severity on a 5-point scale, enabling early treatment and preventing deterioration.

Why is ML more accurate than traditional threshold methods?

Traditional thresholds (e.g., activity > 200% baseline) produce many false positives in stressful situations (heat, regrouping). ML accounts for context: circadian rhythms, seasonality, individual variability. An ensemble of models on activity, rumination, temperature, and milk yield reduces false alarms by 3–5 times. According to a study published in Journal of Dairy Science, using ML improves estrus detection accuracy by 15% compared to threshold methods.

How do we implement the monitoring system?

1. Sensor and infrastructure audit – we check compatibility of ear tags, boluses, milking robots with our platform. Supported sensors: SCR, Allflex, Smaxtec, Moocall, Lely Astronaut.
2. Model calibration for breed and climate – collect baseline data for 2-3 weeks, adjust thresholds and feature engineering. We use PyTorch for time series and LangChain for building RAG pipelines.
3. ML pipeline development – PyTorch, Hugging Face Transformers, ChromaDB for storing embeddings. Apply LoRA adaptation and INT8 quantization to reduce latency.
4. Integration with Farm Management Software (Agrosoft, DairyComp 305, 1С:Ferma) via REST API.
5. Dashboard and alerts – web interface, Telegram/SMS notifications for veterinarians.
6. Staff training – we conduct a 2-day training.

Example code: estrus detection

def detect_estrus(activity_data, cow_id, lookback_days=21):
    """
    Охота = резкий рост активности + падение руминации
    Цикл: 21 день → алерт при аномальном пике активности
    """
    activity_baseline = activity_data.rolling(21 * 24).quantile(0.5)  # медиана за 21 день
    activity_ratio = activity_data / activity_baseline

    rumination = get_rumination_data(cow_id)

    estrus_score = (
        activity_ratio *                            # рост активности
        (1 - rumination / rumination.rolling(7 * 24).mean())  # падение руминации
    )

    # Порог: estrus_score > 1.5 в течение 4+ часов
    estrus_alert = (estrus_score > 1.5).rolling(4).min() > 0
    return estrus_alert, estrus_score

What is included in the project?

• Analytical report – audit of current sensors, infrastructure, and data quality.
• Calibrated ML models – tailored to your breed, climate, and housing type.
• Integration module – REST API for connecting to Farm Management System.
• Dashboard and alert system – web interface, Telegram/SMS.
• Documentation and training – 2-day training for staff, technical documentation.
• Warranty support – 3 months post-project support.

Technical details: mastitis model

def mastitis_risk_score(cow_id, current_features):
    features = {
        'rumination_drop_pct': (current_features['rumination_today'] -
                                 current_features['rumination_7d_mean']) / current_features['rumination_7d_mean'],
        'milk_yield_drop_pct': (current_features['milk_today'] -
                                 current_features['milk_7d_mean']) / current_features['milk_7d_mean'],
        'temp_deviation': current_features['rumen_temp'] - cow_history['temp_baseline'],
        'activity_change': current_features['activity_today'] / current_features['activity_7d_mean']
    }
    return mastitis_model.predict_proba([list(features.values())])[0][1]

The model is based on gradient boosting (CatBoost) and trained on historical farm data. ROC-AUC >0.92.

Group analytics and forecasting

Herd stress index – if 20%+ of cows show a drop in rumination simultaneously → systemic issue (feeding, ventilation, heat). Heat stress – Temperature Humidity Index (THI) > 68 → decreased productivity. ML forecast of milk losses based on weather forecast. Additionally, we build lactation curves for each cow, allowing dry-off and calving planning.

Sensor comparison

Timeline and conditions

Basic system (estrus detection, fever alerts, herd stress) – 4–5 weeks turnkey. Extended ML analytics (mastitis, acidosis, lameness, lactation forecast) – 2–3 months. Cost is calculated individually based on number of heads, sensor types, and required integration. We guarantee prediction accuracy above 90% after calibration.

Contact us for a project assessment – we'll send a demo on your data. Get a free engineer consultation. Over five years of experience in cattle analytics, 15+ implementations on farms from 100 to 5000 heads. We use stacks from SCR, Allflex, Smaxtec, Lely Astronaut. We provide model compliance certificates. Order a trial analysis of your herd's data – we'll show real savings on your numbers.

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.