Eureka! Institutional Repository

In the field of data science a well-known variant of the KNN algorithm is the AdaNN algorithm. The KNN algorithm is a widely used algorithm for many years, which has some advantages and disadvantages. One of its main disadvantages is the fixed number of k, which determines the size of each instance’s neighborhood and therefore affects the final result of instance classification. A basic variant of this algorithm is AdaNN, which sets a different k for each instance of the dataset. The purpose of the thesis is to make an implementation of the basic AdaNN algorithm and to compare the performance of each variant with the AdaNN algorithm and the traditional KNN. Also in this thesis, an investigation and implementation of new variants of the AdaNN algorithm will be implemented in the Python programming language, the methodology of each variant will be analyzed and a comparison will be made with the results of the simple version of the AdaNN algorithm as well as a comparison with KNN.