Cyber attack detection is an important and challenging area of research in the field of information technology. In such a scenario, intruders introduce new mechanisms by applying polymorphic mechanisms in order to escape from the intrusion detection systems. This leads to loss in data and increase in security vulnerabilities. In the past, many soft computing techniques were used from the field of machine learning for enhancing the efficiency of intrusion detection systems (IDSs) in computer networks. Among them, fuzzy logic played a major role for making effective decisions. In addition, the neural networks are also contributing more in this area for training the datasets to form rules which can be used to develop an effective intrusion detection system. In this paper, we propose a new intrusion detection system by combining neural networks with temporal and type-2 fuzzy logic for performing effective classification of the dataset. In addition, a new feature selection algorithm is also proposed in this paper which uses information gain of attributes with fuzzy logic decision making for selecting the optimal number of features from the dataset. This work has been tested by using NSL-KDD dataset and through the experiments conducted in this work it is proved that the proposed system increases the intrusion detection accuracy and reduces the false positive rate when it is compared with other existing systems. Copyright © 2018 Inderscience Enterprises Ltd.