Extraction of Numerical data from Categorical Data Set and Artificial Neural Networks

In this study, the need for numerical data which is required for the machine learning techniques is considered. In order to classify a categorical data by multilayer perceptron, which is one of the well-known artificial neural network structure, the content of the data is digitalized. While the numerical equivalent of the data content is created at categorical level, the protection of the information it carries has been an important issue in terms of achieving effective results and has been studied a lot. Here, a binary-defined transformation is applied to the categorical data. Afterwards, a guide vector is implemented to protect the relation at its own business context. The guide vector implicitly conserves positional relation of the corresponding dataset at the categorical level. Hence, a significant dimensional reduction is accomplished in this new dataset obtained after the digitalization of the original dataset. The classification performance of newly extracted features is validated with multilayer perceptron structure and successful results are observed. In this study, the actual data extracted from the enterprise resource planning (ERP) system is used as the main data source. The underlying methods of modeling and data processing are implemented in MATLAB and Python programming languages.