A Stability Improved Feature Selection Method for Classification of Ship Radiated Noise

Feature selection is an effective method to reduce feature dimensions and avoid overfitting risks in classification problems. Wrapper-based feature selection methods using holdout methods to estimate the accuracy of the classification will lead to instability of the feature selection results and do not meet the needs of practical applications. We propose a feature selection algorithm with improved stability. We use the sequential forward selection algorithm to introduce the confidence interval upper bound for classification accuracy. Experimental results show that this method can meet the requirements of classification accuracy and stability at the same time.