KI-Glossar
Das vollständige Wörterbuch der Künstlichen Intelligenz
Rolling Window
Fixed-size time window that slides over a time series to compute statistics on consecutive subsets of data.
Moving Average
Average calculated over a defined number of consecutive points in a time series, allowing to smooth fluctuations and reveal underlying trends.
Time Series Resampling
Process of changing the sampling frequency of a time series by aggregation (downsampling) or interpolation (upsampling) of data.
Lead Features
Variables created by temporally shifting observations forward, used to predict future values or capture anticipatory trends.
Temporal Aggregation
Technique of consolidating temporal data over predefined periods (hour, day, week) to reduce noise and reveal macroscopic patterns.
Window Functions
Analytical functions performing calculations on a set of data rows relative to the current row within a defined time window.
Cumulative Sum
Temporal aggregate calculating the progressive sum of values from the beginning of the series, useful for detecting growth trends and inflection points.
Time Bucketing
Technique of discretizing time into fixed or variable intervals to group and aggregate temporal data in a structured manner.
Seasonal Decomposition
Analytical method separating a time series into trend, seasonal, and residual components to better understand temporal patterns.
Sliding Window
Fixed-size time window moving step by step over data, used for local analysis and temporal pattern detection.
Temporal Drift
Gradual change in statistical properties of temporal data over time, requiring continuous adaptation of predictive models.
Time-based Features
Variables derived from temporal properties such as hour, day of week, month, or specific periods to capture cyclical patterns.
Rolling Standard Deviation
Volatility measure calculated over a sliding time window, indicating data variation around their moving average.
Temporal Cross-validation
Cross-validation technique respecting chronological order of data to evaluate model performance on realistic temporal predictions.
Expanding Window
Time window whose size gradually increases to include all historical data up to the current observation point.
Temporal Feature Scaling
Normalization of temporal features considering their evolution over time to avoid biases related to variable scales.
Windowed Fourier Transform
Frequency analysis applied to time segments to identify local periodic patterns in time series.