Realization of new optimization-based methods to improve the cryptological properties of traditional random number generators

Data security has become an important strategic element in today's technology where digitalization is increasing rapidly. The fact that modern cyber threats are becoming more sophisticated every day highlights the importance of security measures to be taken in this field. The science of encryption and information security has become an important subject that has been intensively studied. The most important element of encryption science is random numbers and the generators that produce these numbers. For a successful encryption, the output of a random number generator should ensure randomness and reduce predictability. In this thesis, we aim to design an optimization-based architecture for determining the initial conditions that are vital for random number generators. As it is known, the randomness of the random numbers to be obtained from random number generators increases or decreases depending on the initial seed value and the correct selection of the parameters. In this case, optimization algorithms were used to determine the initial conditions. With optimization algorithms, configurations with high randomness and good statistical properties were determined from infinite space. Particle swarm optimization, bat algorithm and modified golden sine algorithm were used to determine the initial conditions and parameters of LFSR, LFG and LCG generators. The randomness of the numbers obtained from the obtained optimization-based architectures was tested with NIST SP 800-22. A displacement box (s-box) design was made with the random numbers generated. In addition, random numbers were obtained using a chaotic pendulum and s-box design and image encryption operations were performed with the random numbers obtained. It is seen that the bits generated with the configurations obtained as a result of the experimental studies successfully passed all randomness tests. When we look at the performance criteria of the generated s-box, it is seen that it gives very successful results. All runs were run until the last iteration and a very high number of results were obtained. As a result, with the results obtained by the proposed optimization-based architecture, an entropy pool consisting of random numbers with high randomness and good statistical properties was created.