In our dataset, without using polynomial features, the accuracy is low. Adding polynomial features is one such technique. In such cases we can engineer new features into the model. Sometimes linear models are unable to capture the dataset properly. pca = PCA(n_components = 4) pca.fit(X) X_pca = pca.transform(X) X = pd.DataFrame(X_pca) Adding polynomial features If we consider first 3 components, it achieve only 83.7% of the covariance, which is less than 95%. We can see that, with the first 4 components we can achieve 96.98% of covariance while the fifth component is contributing only 3.01%, which may lead the model to overfit. correlation_matrix = X.corr() correlation_matrix Let’s calculate the correlation matrix and the heatmap for our features. In that situation we have to drop one of them. A high correlation (positive or negative) between two features means they are highly correlated. We can use the correlation matrix to identify the independent features. Using only the optimal features helps to reduce the complexity of the model and the computational cost. New features that are significant for the predictions can also be included in this step. Feature engineering aids in choosing the optimal features for the model training. Perform Feature EngineeringĪnother essential phase in machine learning is feature engineering because more number of features does not imply more accuracy.
Therefore it is not necessary to apply feature discretization. But in this case the ranges are not huge. Discretization is done when a feature lays on a wide range.
In feature discretization we transform continuous features into discrete features.
0 kommentar(er)
