Cyber-physical　wireless　systems　have　surfaced　as　an　important　data　communication　and　networking　research　area.　It　is　an　emerging　discipline　that　allows　effective　monitoring　and　efficient　real-time　communication　between　the　cyber　and　physical　worlds　by　embedding　computer　software　and　integrating　communication　and　networking　technologies.　Due　to　their　high　reliability,　sensitivity　and　connectivity,　their　security　requirements　are　more　comparable　to　the　Internet　as　they　are　prone　to　various　security　threats　such　as　eavesdropping,　spoofing,　botnets,　man-in-the-middle　attack,　denial　of　service　(DoS)　and　distributed　denial　of　service　(DDoS)　and　impersonation.　Existing　methods　use　physical　layer　authentication　(PLA),　the　most　promising　solution　to　detect　cyber-attacks.　Still,　the　cyber-physical　systems　(CPS)　have　relatively　large　computational　requirements　and　require　more　communication　resources,　thus　making　it　impossible　to　achieve　a　low　latency　target.　These　methods　perform　well　but　only　in　stationary　scenarios.　We　have　extracted　the　relevant　features　from　the　channel　matrices　using　discrete　wavelet　transformation　to　improve　the　computational　time　required　for　data　processing　by　considering　mobile　scenarios.　The　features　are　fed　to　ensemble　learning　algorithms,　such　as　AdaBoost,　LogitBoost　and　Gentle　Boost,　to　classify　data.　The　authentication　of　the　received　signal　is　considered　a　binary　classification　problem.　The　transmitted　data　is　labeled　as　legitimate　information,　and　spoofing　data　is　illegitimate　information.　Therefore,　this　paper　proposes　a　threshold-free　PLA　approach　that　uses　machine　learning　algorithms　to　protect　critical　data　from　spoofing　attacks.　It　detects　the　malicious　data　packets　in　stationary　scenarios　and　detects　them　with　high　accuracy　when　receivers　are　mobile.　The　proposed　model　achieves　better　performance　than　the　existing　approaches　in　terms　of　accuracy　and　computational　time　by　decreasing　the　processing　time.