Software Anomaly Detection Method Based on Artificial Neural Network

A growing range of data sets have been created in recent years; these are used by platforms and software applications and kept in remote access repositories. Datasets are therefore more susceptible to harmful attacks. As a result, network security in data transmission is becoming a more important area of study. One well-known method for safeguarding computer systems is the deployment of intrusion detection systems. This study proposes an artificial intelligence based method for data analysis-based anomaly detection. Methods based on machine learning and rules are mixed together. The right rules are created via a genetic algorithm. Relevant features are extracted using principal component analysis with the goal of enhancing performance. The KDD Cup 1999 dataset is used to empirically validate the suggested procedure, satisfying the criterion of using appropriate data. Using the well-known benchmark dataset, the suggested approach is used to identify and examine four different kinds of attacks: Neptune, Ipsweep, Pod, and Teardrop. During the machine learning phase, the data is categorized into categories of attacks and normal behavior after the features set during the training phase are tested. For the purpose of data analysis, the input data is divided into training and testing sets for an artificial neural network. The first 80% of the data are used to train the neural network, and the remaining 20% are used for testing. The estimated accuracy improves with the number of epochs and is higher for training data and lower for validation test data, according to experimental results. Consequently, the trained model can be retained and used to detect discrepancies. The learnt model is used to the system to compute new input parameters that are not used during training or validation.