نوع مقاله : مقاله پژوهشی
نویسندگان
دانشکده مهندسی برق و کامپیوتر، دانشگاه تبریز
چکیده
کلیدواژهها
عنوان مقاله [English]
نویسندگان [English]
In contemporary times, dealing with enormous data dimensions has emerged as a critical challenge. The "curse of dimensionality" due to high data dimensions poses a continuing challenge to machine learning algorithms. To overcome this problem, many feature selection algorithms have been proposed. However, the presence of imbalanced data in some applications limits the efficiency of most feature selection algorithms and pushes them towards the majority class. Therefore, this study introduces a new bacterial-based evolutionary feature selection algorithm that maintains its efficiency even in the case of imbalanced data. The proposed method employs intelligent mutation and injection operators, which simulate the behavior of bacteria. In these two operators, prior bacteria transfer their best-ranked and prominent features to other bacteria, and the mutation operator replaces those features with the lowest rank. The elitism operator aims to enhance archive members by injecting the best features into other members. The proposed method outperforms conventional methods in terms of feature selection reduction and accuracy enhancement by 43% and 7%, respectively, on the Isolet dataset. The selected features are evaluated using F-measure. The simulation results indicate that the proposed method performs better in selecting the number of features with respect to the algorithm's error compared to state-of-the-art methods.
کلیدواژهها [English]
)