Combined Method of Feature Selection for Machine Learning Models in Solving the Problem of Forecasting

The paper proposes a method for comprehensive modification of the parameter space when synthesizing machine learning models for noise ratio prediction tasks, based on replacing the least relevant group of features with their linear combination. Based on experimental data, the effectiveness of the proposed approach was evaluated and compared with other algorithms in terms of computational complexity and modeling accuracy. It was determined that the proposed method significantly reduces the computational complexity of models at the training stage with a slight increase in error on the test data set.