A Hybrid Algorithm for Constructing an Optimized Set of Informative Features
Annotatsiya
The article emphasizes that extracting informative features from existing datasets is a key step in the process of preparing data for data mining algorithms. This process is aimed at identifying informative features that improve the quality of the model and reduce its complexity. Metaheuristic-evolutionary algorithms have proven their effectiveness in identifying genes, disease signatures, and keywords that provide information in the field of bioinformatics, medical diagnostics, and knowledge mining. The task of the ant colony algorithm is to use a pheromone-based path selection strategy that allows for faster and more accurate approach to the optimal solution in extracting informative features. It is proven that combining ant colony and genetic algorithms into a hybrid algorithm allows using the strengths of both methods to more efficiently select informative features. This hybrid approach allows selecting informative features by increasing the performance of intelligent analysis models and reducing their complexity. To test the developed algorithm, computational experiments were conducted on CDD and ECG datasets used in the diagnosis of cardiovascular diseases. Extraction of informative features from these datasets was carried out using ant and genetic-ant algorithms. For each dataset, classification accuracy was calculated using the Naive Bayes, Decision Tree, KNN, Random Forest and SVM algorithms based on common features and selected informative features, and then the results were compared. The comparison results are presented in the table. Using the developed genetic-ant hybrid algorithm, it is possible to form an informative subset of features in datasets used to solve problems in various fields. The goal is to reduce the size of datasets by compactifying the feature space, thereby reducing the execution time of the algorithm and creating new datasets with higher classification accuracy.