Mobile Ad Hoc Networks (MANETs) are a type of infrastructure-less network; therefore they rely on neighbour cooperation to facilitate routing of data. Such a property leaves them prone to many attacks. When intrusion detection systems are applied to MANETs; high-dimensional traffic data containing largely irrelevant or redundant features results in high computational cost and poor detection rates. Hence, feature selection becomes a crucial precursor to applying any intrusion detection algorithm to MANET environments. GWO (Grey Wolf Optimization) is popular for wrapper-based selection of features due to its simplicity and robust global search characteristics, but was designed for continuous rather than binary selection.In this study we develop an approach which applies an X-shaped transfer function on GWO’s position to generate binary selected features and we guide the algorithm’s initialisation based on Information Gain of the features. Named XTF-GWO, this algorithm was evaluated with network data obtained from ITU-ML5G-PS-006 (2025) repository and coded in MATLAB R2025a, classifying data with Support Vector Machine. Four levels of attack injection were analysed (25%, 50%, 75%, 100%). The experiments conclude that an improved feature exploration and exploitation of features through an X-shaped transfer function and Info Gain initialisation, led XTF-GWO to more discriminated feature subsets than GWO and hence better (on the 4 studied proportions of attack detection) results of detection rate (88.42% vs 82.10% on average, up to 96.82% vs 89.15%), with greater precise and recall values. Keywords: Feature selection, Grey Wolf Optimization, transfer function, intrusion detection, Mobile Ad Hoc Network