PRINCIPLES OF OPTIMIZING RECOGNITION ALGORITHMS BASED ON THE ANALYSIS OF STRONGLY CORRELATED FEATURE SETS
Ключевые слова:
strongly related features, representative features, algorithm efficiency, data optimization, classification, computational complexityАннотация
This article examines the problem of optimizing the feature space in recognition and classification algorithms that work with high-dimensional data. A strong relationship between characteristics increases the complexity of calculations, increases the risk of overtraining the model, and reduces the accuracy of recognition. Taking this into account, the article proposes a two-stage method for optimizing the feature space: in the first stage, the degree of correlation between features was assessed using various metrics (Euclidean distance, Pearson's correlation coefficient, entropy), which were divided into clusters; in the second stage, the most informative features were selected from each cluster. The scientific novelty of the article lies in ensuring the selection of representative features for each subset, combining correlation criteria and information capacity when structuring the feature space, which allows for increased accuracy while maintaining the computational efficiency of the model
