Automated organization of interaction between modules of information systems based on neural network data channels

The automation of the process of information systems construction is an important and urgent problem, as it allows reducing the negative influence of a person during decision making and developing software, releasing additional time and material resources to solve more complex and creative problems. Most modern information systems are developed on a modular base; therefore, a significant design stage is the implementation of links between system components. The purpose of the study is to automate the organization of intermodular interaction in information systems, which will reduce the complexity, time and cost of this implementing process. In order to achieve the result, a method is proposed for the organization of interaction between modules of information systems based on neural network data channels, realized within the general concept of a neural network architecture. The structure of neural network channels, the principles of their functioning, theoretical substantiation, mathematical and algorithmic support and area of application are considered in detail. A classification of neural network channels is presented, based on two of their characteristics: categories and degrees. As a result of the conducted research, the practical implementation of two neural network data channels is realized (transmission and adaptation), the structure of the program code, the used tools and libraries are analyzed. Based on a set of metrics for the complexity of the program code (Halstead, Jilb), the estimation of the computational complexity of algorithms, time and material costs for implementation, a comparative analysis of neural network data transmission channels and adaptation with classical approaches in the form of a set of network data transmission protocol and the required algorithmic support for data processing is carried out. The obtained experimental results confirm the lower complexity of neural network channels (reduction by at least 20% according to Halstead metrics and cyclometric complexity), reduction in time (by 12–32%) and cost (by 36–63%) of implementation and increase in the accuracy of the problem solving (by 11.8–15.5%). This demonstrates the effectiveness of using neural network data channels to automate the organization of intermodular interaction in information systems.