In this study, feature selection methods based on the new Caledonian crow learning algorithm has been introduced. In the proposed algorithms, in the first stage, the best features related to COVID-19 disease are selected by the crow learning algorithm. Coronavirus (COVIDE-19) disease using as training input to the artificial neural network. Experiments on the COVID-19 disease dataset in a Brazilian hospital show that the crow learning algorithm reduces the feature selection objective function by iteration. Decreasing the feature selection function is due to reducing the error of classifying infected people as healthy and reducing the number of features. The experimental results show that the accuracy, sensitivity, precision, and F1 of the proposed method for COVID-19 patients diagnosing are 94.31%, 94.15%, 94.38%, and 94.27%, respectively. The proposed method for identifying COVID-19 patients is more accurate than ANN, CNN, CNNLSTM, CNNRNN, LSTM, and RNN methods.