The most appealing characteristic o

The most appealing characteristic of DL is the ability to extract high-level features to leverage the large number of unlabeled instances. Based on the extracted features, classifiers such as support vector machine (SVM) and random forest (RF) are used to build the prediction model. However, while a single classification model performs well, fusing multiple models through ensemble learning improves performance. An ensemble learning method is a meta-algorithm that trains several baseline models and combines them into a single predictive model. Apart from improved classification, ensemble methods perform well in problems involving noisy and imbalanced datasets [38]. Classification strengths of individual base classifiers selected for construction of the overall ensemble model lead to more accurate predictive performance. In 2011, a method named RPISeq [39] extracted 3-mer and 4-mer sequence features to train RF and SVM models for prediction of protein-RNA interaction. Then, Wang et al. [40] presented a model that predicted interactions between proteins and RNAs based on Naive Bayes (NB) and an extended NB classifier. In 2016, Pan et al. developed IPMiner, a sequence-based method for predicting lncRNA-protein interactions based on stacked autoencoder [41]. Yi et al. proposed RPI-SAN for lncRNA-protein interaction based on stacked autoencoder and RF [42]. lncLocator [43] predicted lncRNA subcellular localizations based on an ensemble of SVM and RF classifiers. A recent tool termed HLPI-Ensemble (human lncRNA-protein interaction) was proposed to predict human lncRNA-protein interactions based on SVM, extreme gradient boost (XGB), and RF [44]. For cancer prediction, a model based on DL with an ensemble of three classifiers was recently proposed by Xiao et al. [45].