Time Series Forecasting at Scale: From ARIMA to LightGBM

Method Selection

Scenario	Best Method
Few series, clean data	Prophet, SARIMA
Many series (1000+)	LightGBM with lag features
Complex seasonality	N-BEATS, TFT
Real-time, low latency	Simple exponential smoothing

LightGBM for Time Series

def make_features(df, lags=[1,7,14,28], windows=[7,14,28]):
    for lag in lags:
        df[f'lag_{lag}'] = df['target'].shift(lag)
    for w in windows:
        df[f'rolling_mean_{w}'] = df['target'].shift(1).rolling(w).mean()
        df[f'rolling_std_{w}'] = df['target'].shift(1).rolling(w).std()
    df['day_of_week'] = df.index.dayofweek
    df['month'] = df.index.month
    df['is_weekend'] = (df.index.dayofweek >= 5).astype(int)
    return df

Backtesting Framework

Walk-forward validation: train on [0..t], predict [t+1..t+h], slide forward.