In　the　digital　transformation　era,　the　Internet　of　Things　(IoT)　has　become　integral,　propelling　innovation　yet　exposing　cybersecurity　vulnerabilities　and　compromising　system　integrity.　The　complexity　and　diversity　of　IoT　environments　render　traditional　security　measures　inadequate　against　the　evolving　nature　of　cyber　threats,　especially　those　targeting　wireless　sensor　networks　(WSNs).　This　gap　underscores　the　need　for　solutions　that　adapt　alongside　emerging　threats　to　enhance　IoT　ecosystem　resilience.　Our　research　introduces　a　pioneering　intrusion　detection　system　(IDS)　combining　reinforcement　learning　(RL)　and　artificial　neural　networks　(ANNs).　Using　a　Q-learning　agent,　this　system　interacts　with　various　IoT　datasets　to　iteratively　select　and　refine　attack　features　for　classification,　enabling　the　identification　of　key　representations　that　increase　detection　accuracy.　The　model　can　effectively　identify　known　and　emerging　threats　through　an　automated　feature　selection　process　and　continuous　adaptation.　Extensive　testing　demonstrated　robust　performance　under　dynamic　network　conditions,　positioning　this　approach　as　a　scalable　and　resilient　defense　for　large-scale　IoT　infrastructures.　When　tested　on　real-world　IoT　datasets　across　multiple　contexts　and　threats,　the　proposed　model　achieved　over　99%　accuracy　while　maintaining　a　false-positive　rate　of　less　than　5　%,　surpassing　existing　solutions　and　establishing　a　new　standard　for　adaptive　cybersecurity.