Context-Aware Ene rgy Enha ncements for Smart Mobile Device s

Abs trac t— Within the pas t decade , mobile computing has morphed into a principa l form of human communica tion, bus iness , and socia l inte raction. Unfortuna te ly, the energy demands of newer ambient inte lligence and collabora tive technologies on mobile devices have grea tly overwhelmed modern energy s torage abilities . This paper proposes severa l nove l techniques tha t exploit spa tiotempora l and device context to predict device wire less da ta and loca tion inte rface configura tions tha t can optimize energy consumption in mobile devices . These techniques , which include variants of linear discriminant ana lys is , linear logis tic regress ion, non-linear logis tic regress ion with neura l ne tworks , k-neares t ne ighbor, and support vector machines are explored and compared on synthe tic and user traces from rea l-world usage s tudies . The experimenta l results show tha t up to 90% success ful prediction is poss ible with neura l ne tworks and k-neares t ne ighbor a lgorithms , improving upon prediction s tra tegies in prior work by approximate ly 50%. Further, an average improvement of 24% energy savings is achieved compared to s ta te -of-the -art prior work on energy-efficient loca tionsens ing.