AI 용어집
인공지능 완전 사전
Sliding Window
Feature engineering method that segments a time series into fixed-size subsequences that move one time step at a time to create training samples.
STL Decomposition (Seasonal-Trend decomposition using Loess)
Robust algorithm that decomposes a time series into three components: trend, seasonality, and residual, facilitating analysis and modeling for predictive maintenance.
Temporal Differentiation
Stationarization transformation that calculates the difference between successive observations to eliminate trend and make the time series more stable for modeling.
Anomaly Detection by Isolation Forest
Cleaning algorithm that isolates observations by building random decision trees, where anomalies are identified by their shorter path to a leaf.
Temporal Min-Max Normalization
Scaling method that transforms time series values into a fixed range [0, 1] using the minimum and maximum of the training window to preserve temporal relationships.
Lag Feature Engineering
Creation of explanatory variables based on lagged values of the original time series to capture temporal dependencies and autocorrelation.
Temporal Aggregation by Moving Average
Noise reduction technique that calculates the average of observations over a sliding time window to smooth short-term fluctuations and reveal underlying trends.
Temporal Z-Score Standardization
Normalization that subtracts the mean and divides by the standard deviation calculated over a time window to center and scale the data, making models less sensitive to scales.
Cyclic Encoding of Temporal Features
Transformation that represents cyclic temporal features (hour, day, month) by their sine and cosine to preserve the circular nature of time in models.
Temporal KNN Imputation
Method for filling missing values that uses the k nearest neighbors in time to estimate the missing value, based on the similarity of adjacent sequences.
Butterworth Low-Pass Filtering
Denoising technique that eliminates high frequencies from a time series while preserving low frequencies, using a filter with a flat frequency response in the passband.
Temporal Resampling
Process that modifies the sampling frequency of a time series through upsampling or downsampling to align data on a uniform temporal grid.
Change Point Detection
Algorithm that identifies moments when the statistical properties of a time series change significantly, crucial for detecting transitions between operational states.
Wavelet Transform
Decomposition technique that analyzes a time series at different frequencies and temporal resolutions, allowing isolation of local features for fault detection.
Creation of Rolling Statistical Features
Generation of descriptive variables calculated on sliding temporal windows such as variance, skewness, or kurtosis to capture the local dynamics of signals.
Spline Interpolation Imputation
Advanced method for filling missing values that uses piecewise polynomials to create a smooth curve passing through known data points.