An ensemble approach to predict local lymph-node metastasis in cancer based on SMOTE strategy

Local lymph node metastasis is a standard mode of tumor metastasis. Accurate prediction of local lymph node metastasis in cancer patients can help select appropriate treat-ment. Long non-coding RNA (LncRNA) has been proved to play an essential role in cancer prediction. According to the characteristics of such data, problems such as data imbal-ance, high dimensionality, and small sample size may arise in the analysis involved. Besides, considering a large amount of adequate information in LncRNA expression profiles and the correlation between different features, the commonly used data dimensionality reduction method may not retain the expression profile information well. In this research, taking complete account of data imbalance, a comprehensive feature extraction method is proposed by combining an ensemble classification strategy with Synthetic Minority Oversampling Technique (SMOTE). The biometric selection and linear discriminant analysis were used to reduce the feature dimension and enhance computational complexity. Some comparative experiments were conducted on three tissue-specific cancer datasets, and the performance validated the effectiveness of the VotMeta in terms of accuracy.