An Algorithm of Robust Online Extreme Learning Machine for Dynamic Imbalanced Datasets

With the coming of big data age, dynamic data has gradually appeared in various application fields, such as safety monitoring, financial forecasting, and medical diagnostics. Although existing knowledge discovery and data mining techniques have shown great success in many real-world applications, dynamic data has the features of imbalance and instability of data classes, the dynamic change of data volume, which makes it difficult for the classification of dynamic data. To solve these problems, in this paper a robust weighed online sequential extreme learning machine algorithm (RWOSELM) based on the online sequential extreme learning machine algorithm (OSELM) is presented. RWOSELM generates the local dynamic weighted matrix with the help of cost sensitive learning theory, thereby it optimizes the empirical risk of the classification model. Meanwhile, RWOSELM takes the data distribution changes which are caused by temporal properties change of dynamic data into consideration, thus it introduces the forgetting factor to enhance the sensitivity of the classifier to the change of data distribution. The method is able to deal with the data with imbalanced class distribution, and maintains the good robust on dynamic data. This paper tests on 24 datasets with different distribution, and the results show that RWOSELM gets good results on imbalanced dynamic dataset.