Page 13 - Special Topic Session (STS) - Volume 4
P. 13

STS556 Mohd Bakri A. et al.
            is compound smoother.  compound smoother is known as a powerful tool to
            smooth a data series without excessively disrupting the details of a data series.
            Despite this good traits, the compound smoother does not respond well to
            oscillated  trend,  Tothmeresz  and  Erdei  (1995)  and  Jin  and  Xu  (2013).  The
            number  observations  of  compound  smoother  should  be  at  least  seven,
            otherwise  it  will  converge  to  constant  root,  Janosky  et  al.  (1997).  The
            Velleman's compound smoother as indicated by Sargent and Bedford (2010)
            has  been  revised  when  possible  combinations  of  multiple  step  of  running
            median, Hanning and resmooth the rough are tested out. Improvement on the
            existing compound smoother in comparison has yet been explored.

            2.  Methodology
                The 4253HT is one of the non-linear smoothing techniques that combined
            running median, weighted moving average and re-smoothed the rough. This
            technique  was  first  introduces  by  (Tukey,  1977)  and  described  detail  in
            different version by (Velleman et al., 1981). Let X be a doubly-infinite sequence
            of  real  data X ,...,  X , X , X ,..., X t+ n  .  A  smoother  M  is  defined  as  an
                                       t
                                          t+1
                                   t−1
                            t−
                             n
            algorithm that works on X to generate a new series M (X t ) , smoothed values.
            The algorithm of 4253HT is as follows:
            Step 1: Perform running median of span size two
                                         M 1 (X t ) =  median (X  − t  2 , X  1 − t  , X t , X  + t  1 )          (1)

            Step 2: Re-centered the equation (1)
                                         M  (X  ) =  median (M  (X  ),M  (X  ) )           (2)
                                2  t            1  t   1    1 + t

            Step 3: Next, equation (2) are smoothed again by applying running median of
            span size three
                          M 3 (X t ) =  median (M 2 (X  − t  2 ),M 2 (X  1 − t  ),M 2 (X t ),M 2 (X  1 + t  ),M 2 (X  + t  2  ) )        (3)

            Step 4: Perform running median of span size three
                                 M 4 (X t ) =  median (M 3 (X  1 − t  ),M 3 (X t ),M 3 (X  1 + t  ) )          (4)

                                                                                      1
                                                                                  1
            Step 5: Apply weighted moving average or Hanning with coefficients  ,
                                                                                  4   2
                 1
            and
                 4
                                  1           1          1
                                 M  (X  ) =  M  (X  ) +  M  (X  ) +  M  (X  )                   (5)
                           5  t       4   − t  1  4  t       4  + t  1
                                  4           2          4
            Step 6: Re-smooth the rough and added to the smoothed values in (5)
                                         M 6 (X t ) =  M 5 (X t ) M+  5 X − M 5 (X t   )                  (6)
                                                     t



                                                                 2 | I S I   W S C   2 0 1 9
   8   9   10   11   12   13   14   15   16   17   18