Abstract:

This study proposed an integrated neuropeptide prediction algorithm. This study integrated nine feature descriptors and five machine learning algorithms in order to generate 45 baseline learning models for predictive training of neuropeptides. The first layer performs feature selection on these 45 baseline models to select the features with good performance. While the second layer selects eight basic learning models based on the accuracy of the baseline model and the sum of Pearson correlation coefficients. The third layer inputs the output of these learners into logical regression， Extreme Gradient Boosting （XGBoost）， and other classifiers for the final step selection to train the final model， and uses the output as the final prediction result. The final accuracy on the test dataset is 0.9169， which is higher than existing models.

Key words: computer application technology, neuropeptide, feature selection, machine learning, stacking method