Page 82 - Contributed Paper Session (CPS) - Volume 5
P. 82

CPS1060 Taik Guan T. et al.
                  growth by about 1.38% [1]. According to the International Telecommunication
                  Union’s  (ITU)  2012  report,  a  10%  increase  in  broadband  penetration  will
                  contribute to a 0.7% increase in Malaysia’s GDP [2]. Many researches have
                  addressed the different magnitudes of the positive impacts of BS, depending
                  on the current economic situations of those geographic areas. However, only
                  the  developing  and  developed  economies  are  normally  assumed  to  be
                  commercially viable for private investments. The underdeveloped economies
                  require legislative support for broadband development because quick profits
                  are  unattainable  in  these  areas.  Collectively,  the  various  research  results
                  provide  confidence  to  the  privately-owned  telco  to  provide  broadband
                  services  in  urban  and  selected  suburban  but  not  rural  areas.  Furthermore,
                  there  are  no  empirical  models  made  available  for  the  telco  to  predict  the
                  economic potential of certain geographic areas so that they can prioritize their
                  network investments in promising rural areas. As a result, the telco continues
                  using  the  ROI  model  to  strategize  their  network  investment  plans  and  to
                  deploy their BS in urban or suburban areas only.
                      This paper aims to provide a framework of machine learning technique
                  (MLT)  which  can  help  the  telco  and  government  agencies  to  predict  if  a
                  selected  geographic  area  has  the  socioeconomic  potential  for  broadband
                  investments. The proposed machine learning technique is an SVM which is a
                  classifier that predicts the socioeconomic potential in correspondence to the
                  local  features  or  characteristics  of  a  geographic  area.  In  this  framework,  a
                  curve-fitting technique will be applied to formulate the empirical model by
                  using prototyping data from the World Bank databank. The empirical model
                  is then used as a fitness function for a GA to generate large virtual data to
                  train, validate and test the SVM. Real-life field data for geographic areas in
                  Malaysia  is  then  provided  to  the  SVM  to  predict  which  areas  have  the
                  socioeconomic  potential  for  broadband  investment.  The  curve-fitting
                  technique  is  a  statistical  model  to  formulate  the  empirical  model  by
                  establishing  the  interdependency  of  the  geographical  features  with  the
                  socioeconomic  status  of  geographic  areas.  The  curve-fitting  examines  the
                  relationship  between  given  sets  of  independent  and  dependent  variable
                  features.  The  fitted curves  obtained can  be  used  in  data  visualization, and
                  finding relationships between two or more variable features [3].
                      The data on local features of rural areas are lacking or difficult to obtain.
                  Hence data on local features of countries from the World Bank database is
                  applied  in  this  research.  The  availability  of  data  from  the  World  Bank’s
                  database is limited by the number of countries worldwide. The few hundred
                  data sets in the World Bank’s database provide a small sample available to
                  train and optimize the accuracy of the SVM. The limited sample size will affect
                  the  accuracy  of  machine  learning.  Large  virtual  samples  are  essential  to
                  address  the  issue  of  insufficient  raw  data,  which  will  help  overcome  the

                                                                      71 | I S I   W S C   2 0 1 9
   77   78   79   80   81   82   83   84   85   86   87