SEARCH RESULT

In today's world, with the rapid increase in digitalization, data security has been of critical importance for both individuals and organizations. In this context, encryption methods are widely used to protect sensitive information and stand out as a fundamental tool in ensuring data security. Encryption functions as an important protection mechanism for information security by guaranteeing that data is only accessible to authorized users. Together with the impact and performance of advanced analytical techniques such as data mining and machine learning on encrypted datasets, constitute one of the focal points of current research. This study aims to evaluate the effects of original datasets and encrypted forms of these datasets using symmetric encryption algorithms on the performance metrics of models trained with artificial neural networks through a comparative analysis method. In the scope of the study, four datasets with different characteristics, including Breast Cancer, Hepatitis, Iris, and Study Depression have been utilized. In the initial phase of the study, the original versions of the data sets used were processed through the relevant data mining preprocessing steps, trained with artificial neural network models, and their performances were evaluated through various metrics. Afterward, the datasets were encrypted using AES (Advanced Encryption Standard), DES (Data Encryption Standard) and RC4 (Rivest Cipher 4) of the symmetric encryption algorithms. The same data mining preprocessing steps have also been applied to the encrypted data sets, and performance analyses have been conducted by training with artificial neural network models. Lastly, the performance results of the artificial neural network for the original and symmetrically encrypted datasets have been analyzed comparatively for each encryption algorithm separately. The findings put forward indicate that the trainings carried out with artificial neural networks after the encryption of the data exhibited a performance very similar to the trainings carried out with the original data. Consequently, this study shows that while symmetric encryption algorithms ensure data security, simultaneously similar performance results can be obtained with the original data in artificial neural networks-based model training. Even though the encryption processes carried out using AES, DES, and RC4 algorithms cause minor acceptable changes in model performance, they do not affect the overall validity of the results. While preserving the performance of artificial neural network training, ensuring data security stands out as an important balancing factor. This development encourages the integration of data security with neural network applications, contributing both to the widespread adoption of these technologies in high-security areas and to the more efficient realization of secure data processing

International Conference on Advanced Technologies, Computer Engineering and Science
ICATCES

Rabia Günbaş Merve Yılmaz Mustafa Servet KIRAN