ABSTRACT
Software applications are becoming more complex and larger than ever before. Such features make avoiding software vulnerabilities extremely challenging. It is also feasible to predict the number of errors in software modules automatically, which can aid developers in appropriately distributing resource limits. Various strategies for detecting and correcting such flaws at a low cost have been proposed. However, these methods’ performance has to be significantly improved. As a result, employing sophisticated learning approaches, we provide a new methodology to estimate the frequency of errors in information systems. To begin, we prepare a public data collection that includes log processing and data standardisation. Second, we prepare data feedback for the profound learning model by modelling data. Finally, we send the modelling data to a deep neural network algorithm that was created expressly to predict the amount of faults. Two well-known data sets are used to test the proposed approach. The assessment’s findings suggest that the proposed strategy is viable and can improve modern approaches. The proposed strategy, on average, reduces the average square error by over 14% and increases the squared correlation coefficient by over 8%.
