Course Content
Data Anomaly Detection
2. Statistical Methods in Anomaly Detection
Data Anomaly Detection
Regularisation
Regularization is commonly employed when dealing with anomalies to mitigate their undue impact on predictive models. While regularization may not directly identify outliers, its primary role is to reduce the influence of outliers on the model's results.
Instead of explicitly detecting outliers, it focuses on making the model more robust and less sensitive to extreme data points.
Regularisation types
| L1 Regularization (Lasso) | L2 Regularization (Ridge) | Dropout Regularization | |
|---|---|---|---|
| Purpose | L1 regularization adds a penalty term to the loss function based on the absolute values of the model's coefficients. It encourages some coefficients to become exactly zero, effectively performing feature selection. | L2 regularization adds a penalty term to the loss function based on the squares of the model's coefficients. It tends to keep all feature coefficients small but doesn't force them to be exactly zero. | Dropout is a regularization technique commonly used in neural networks. During training, dropout randomly deactivates a fraction of neurons in each layer. This helps prevent overfitting by reducing the reliance on specific neurons or features. |
| Impact on Anomalies | L1 regularization can help make the model more robust to anomalies by reducing the impact of less important features. Anomalies may have less influence on the final model because some feature coefficients are pushed to zero. However, it may not completely eliminate the effect of outliers. | Ridge regularization can make the model less sensitive to anomalies by keeping feature coefficients small. It helps prevent extreme values in coefficients caused by outliers, leading to a more stable model. | Dropout can improve the model's robustness to anomalies by preventing it from becoming overly dependent on specific data points, including outliers. It encourages the network to learn more generalized representations, which can help in handling unexpected or noisy data. |
Everything was clear?
Course Content
Data Anomaly Detection
2. Statistical Methods in Anomaly Detection
Data Anomaly Detection
Regularisation
Regularization is commonly employed when dealing with anomalies to mitigate their undue impact on predictive models. While regularization may not directly identify outliers, its primary role is to reduce the influence of outliers on the model's results.
Instead of explicitly detecting outliers, it focuses on making the model more robust and less sensitive to extreme data points.
Regularisation types
| L1 Regularization (Lasso) | L2 Regularization (Ridge) | Dropout Regularization | |
|---|---|---|---|
| Purpose | L1 regularization adds a penalty term to the loss function based on the absolute values of the model's coefficients. It encourages some coefficients to become exactly zero, effectively performing feature selection. | L2 regularization adds a penalty term to the loss function based on the squares of the model's coefficients. It tends to keep all feature coefficients small but doesn't force them to be exactly zero. | Dropout is a regularization technique commonly used in neural networks. During training, dropout randomly deactivates a fraction of neurons in each layer. This helps prevent overfitting by reducing the reliance on specific neurons or features. |
| Impact on Anomalies | L1 regularization can help make the model more robust to anomalies by reducing the impact of less important features. Anomalies may have less influence on the final model because some feature coefficients are pushed to zero. However, it may not completely eliminate the effect of outliers. | Ridge regularization can make the model less sensitive to anomalies by keeping feature coefficients small. It helps prevent extreme values in coefficients caused by outliers, leading to a more stable model. | Dropout can improve the model's robustness to anomalies by preventing it from becoming overly dependent on specific data points, including outliers. It encourages the network to learn more generalized representations, which can help in handling unexpected or noisy data. |
Everything was clear?
