Transforming Wearable Data into Clinical Biomarkers with AI

Professional AI Processing of Wearable Device Data

Imagine your wearable device collecting terabytes of raw data, but clinically meaningful metrics remain elusive. Motion artifacts, baseline drift, sensor noise—all make raw signals unusable for diagnosis. We build pipelines that transform raw PPG, accelerometer, ECG, and CGM readings into ready-to-use biomarkers. Clients come to us with this problem, and we solve it with adaptive filters and ML models. These AI health applications are transforming patient monitoring.

Devices and Data We Work With

Why Raw Signals Can't Be Used Directly?

The main issues: motion artifacts (PPG during walking is distorted by 30-50%), baseline drift, missing R-peaks on ECG, and battery depletion artifacts on CGM. Without cleaning, metrics are unreliable. For example, a standard bandpass filter reduces heart rate error by only 10-15%, while our adaptive method reduces it by 2.5 times more.

Signal Processing Modules

PPG Artifact Removal — We use the accelerometer as a reference signal and adaptive filters. The TROIKA and JOSS algorithms reduce average heart rate error from 8 to 3 BPM in test datasets (IEEE SP Cup). Example bandpass filtering for PPG:

from scipy.signal import butter, filtfilt, find_peaks
import numpy as np

def ppg_to_hr(ppg_signal, sampling_rate=25):
    nyq = sampling_rate / 2
    low, high = 0.5 / nyq, 4.0 / nyq
    b, a = butter(4, [low, high], btype='band')
    filtered = filtfilt(b, a, ppg_signal)
    peaks, _ = find_peaks(filtered, distance=sampling_rate * 0.4)
    rr_intervals_sec = np.diff(peaks) / sampling_rate
    hr_bpm = 60 / np.mean(rr_intervals_sec)
    return hr_bpm, rr_intervals_sec

Heart Rate Variability (HRV) Analysis — From clean R-R intervals, we compute time-domain and frequency-domain metrics:

def compute_hrv_metrics(rr_intervals_ms):
    rr = np.array(rr_intervals_ms)
    return {
        'rmssd': np.sqrt(np.mean(np.diff(rr)**2)),
        'sdnn': np.std(rr),
        'pnn50': np.mean(np.abs(np.diff(rr)) > 50),
        'mean_rr': np.mean(rr),
        'mean_hr': 60000 / np.mean(rr)
    }

Heart rate variability is the gold standard for assessing recovery after exercise. RMSSD — marker of parasympathetic activity, SDNN — overall indicator of autonomic tone. Our pipelines compute these in real time on the edge.

Activity Classification with IMU — IMU (accelerometer + gyroscope) provides three-axis data. We extract features in a sliding window: mean, standard deviation, 95th percentile, energy, SMA, zero-crossing rate, dominant frequency, spectral entropy. A RandomForest classifier is trained on 10,000 windows from open datasets (WISDM, UCI HAR). Accuracy: 94% on 6 classes (rest, walk, run, bike, climb, fall).

Sleep Staging ML — Using accelerometer + heart rate, we achieve 82% agreement with PSG across 4 classes (Wake/Light/Deep/REM). This is 20% more accurate than commercial alternatives, matching expensive clinical systems at lower cost.

Continuous Glucose Monitoring — CGM provides a glucose reading every 5 minutes. We extract time-in-range, coefficient of variation, GMI, and detect postprandial peaks. A hypoglycemia prediction model (based on LSTM) forecasts events 30 minutes ahead with AUC 0.89.

Each model comes with a model card: training data description, class distribution, accuracy/recall/precision per class, metrics for different subgroups (gender, age). We use Weights & Biases for experiment tracking, MLflow for versioning. We retrain regularly on new data.

Signal Processing Pipeline Stages

What's Included in Our Work

1. Consultation on sensor selection, sampling rates, storage protocol.
2. Signal processing pipeline in Python (SciPy, custom C++ for edge).
3. ML models (PyTorch for wearable devices / scikit-learn) for classification, regression, prediction.
4. Deployment on edge (Triton, ONNX Runtime) or cloud (Kubeflow).
5. Documentation: model card, feature description, integration guide.
6. Training your team to work with models.

Pricing and Timelines

Estimated timelines depend on complexity and number of sensors.

• Basic processing of one signal (PPG or IMU) + dashboard: starting at $15,000, 4–5 weeks.
• Full stack (PPG + IMU + sleep + CGM): starting at $50,000, 2–3 months.

By using our pipeline, you can save up to $5,000 compared to in-house development. Cost is calculated individually. Get a consultation — we will assess your project. Our team includes experienced ML engineers for wearables, with over 5 years of experience in biomedical signal processing.

Confidentiality

Contact us for a detailed discussion of your project.

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.