Distributed network attacks, also known as distributed denial of service (DDoS) are a major threat and problem for internet security. DDoS is an attack on a network aiming to disable server resources. These attacks increase every year with the current state of the COVID-19 pandemic. One of countermeasures to minimize the DDoS impact is the intrusion detection system (IDS) command. IDS techniques are currently still employing traditional methods so that they have many limitations compared to techniques and tools used by attackers because traditional IDS methods only use signature-based detection or anomaly-based detection models which cause many errors. Network data packet traffic has a complex nature, both in terms of sizes and sources. This research utilized the ability of artificial neural network (ANN) to detect normal attacks or DDoS. A classification technique with ANN method is a solution to these issues. Based on the shortcomings of the traditional IDS, this study aims to detect DDoS attacks using feeder machine learning-based feature engineering techniques to improve the IDS development. Using the UNSW-NB15 dataset with the ANN method, this study also aims to analyze and obtain training function combinations and the best hidden layer architectures of ANNs to solve the detection and classification problems of DDoS packets in computer networks. As a result, the training function combinations and hidden layer architectures of the ANN can provide a high level of DDoS recognition accuracy. Based on experiments conducted with three schemes and an ANN schema architecture technique with eight features as input, the highest accuracy was 98.22%. Feature selection plays an essential role in detection result accuracies and machine learning performances in classification problems.