Skip to main content
. 2022 Jul 17;23(14):7877. doi: 10.3390/ijms23147877

Figure 2.

Figure 2

The accuracy of independent tests for predicting individual and fused features of bitter peptides. Three individual features, SSA, UniRep, and BiLSTM, three double fusion features, SSA + UniRep, SSA + BiLSTM, and UniRep + BiLSTM, and a triple fused feature, SSA + UniRep + BiLSTM, were tested with three distinct machine learning algorithms. The same color is used to signify the same features. The accuracy of the individual or fused feature/machine learning algorithm combinations is ranked from highest to lowest. The accuracy of one of the triple fusion features and four fused double features outperformed the best performing individual feature. In contrast, the four least accurate predictors were individual feature models, indicating the superiority of fused features. The accuracy of the SSA + UniRep + BiLSTM-LGBM) combination was 0.898. The best performing SSA-SVM and BiLSTM-SVM individual features had an accuracy of 0.883. In summary, combining different feature information sets helped to improve the predictive performance of the model. Please note SSA-SVM means that the SVM model with SSA feature vectors as input while and SSA + UniRep + BiLSTM-LGBM means that the LGMB model with the combination of SSA, UniRep and BiLSTM features as input. Other similar labels have similar meanings.