M.A. Thesis:

Classification of incomplete data based on belief functions and K-nearest neighbors

One of the issues in the design of data can add to the complexity of the analysis is the lack of data. Determining these values is of paramount importance in data processing and machine learning. Therefore, in processing processes, it is attempted to use automated techniques to recover missing values. In this thesis, a hybrid approach is introduced Which is based on K-Nearest neighborhood And the belief functions are based on the Dempster- Shafer theory And the first problem that we are examining is solving the uncertainty problem. Using estimates based on the K nearest neighbor, the outputs are estimated. In the following, the estimated results are improved by K's best neighboring neighborhood using the belief function and the mass function selection. The weight factor is the result of the classification of the classes, which can be chosen by the total weight of K and Consider the vectors of the estimation to share the lost values. In addition to these cases, the classification of the result is repeated in series for several types of data that inherently have lost values. For benchmarking of the algorithm, six data have been used in two stages of classification of data along with missing values and data with no loss values. In the first stage, data on kidney disease, cervical cancer and hepatitis have been analyzed, with different percentages of data loss. Finally, the accuracy of the test stage was obtained at 96.55%, 92.23% and 83.71%, respectively. In the second step, by making the lost values in the data of Glass, Thyroid and Wine disease from 5% to 50%, the recovered outputs were determined, and the absolute error and mean square error were calculated. For all three data, these two values had fewer errors compared to techniques like the most similar element and K nearest neighboring which will be reported separately. Applying this combination method can greatly improve the accuracy of classification and predict outcomes that have little or no error compared to actual outputs

Keywords: Missing values, K- Nearest Neighbor, Belief function, Dempster–Shafer Theory and Uncertainity