The paper addresses basic machine learning problems that can be solved using various classification models, including the simplest perceptron neural network. Such methods include a model based on the logic of a classification process that assigns a class based on the class of its closest neighbors. This model is called the k-nearest neighbor algorithm. Software implementations of the nearest neighbor method are proposed that identify the nearest training data point to the new example and then classify it in exactly the same way. In this case, the training data separation line is generated using the Voronoi diagram. The analysis of the results shows that the nearest-neighbor method is much more efficient than the perceptron. In this case, false classifications can be avoided by starting with the k-nearest neighbors’ classification. The article presents a classification training example that uses the nearest-neighbor method for more than two classes, based on data from an autonomous robot. Classification accuracy can be improved by approximating the control function.