Fig. 4: Overall analysis of the correlation between predicted and observed training effects, including an analysis of feature importance. | npj Science of Learning

Fig. 4: Overall analysis of the correlation between predicted and observed training effects, including an analysis of feature importance.

From: Investigating lexical categorization in reading based on joint diagnostic and training approaches for language learners

Fig. 4

A Distribution of predicted vs. observed correlations resulting from leave-one-out-cross-validation across all variations of the machine learning pipeline. B The most frequently selected features among the 20 feature extraction models and nine different feature selection procedures. As features are selected 75 times for each participant due to the leave-one-out-cross-validation, the maximal frequency is 13,500. Only the top 20 features with the highest occurrences are displayed here. The green color highlights the LCM-related features.

Back to article page