ABSTRACT

Hasil produksi tanaman padi lima tahun terakhir di Kabupaten Grobogan mengalami
naik turun setiap tahunnya. Jika produksi padi menurun maka akan meyebabkan persediaan pangan
menjadi tidak stabil. Pemerintah Daerah Kabupaten Grobogan harus mengupayakan kebijakan
agar produksi padi dapat selalu terpenuhi. Salah satu caranya adalah melihat perkiraan produksi
padi pada tahun berikutnya dapat dilakukan dengan melakukan peramalan.. Konsep peramalan
Fuzzy Time Series diperkenalkan oleh Song & Chissom pada tahun 1993. Sistem peramalan
dengan Fuzzy Time Series panjang interval telah ditentukan oleh awal proses peramalan. Salah
satu metode peramalan dengan penentuan panjang interval yang efektif sehingga mampu
memberikan hasil ramalan dengan tingkat akurasi yang baik adalah model Average Based Fuzzy
Time Series yang diperkenalkan oleh Sun Xihao & Li Yimin pada tahun 2008.

Hasil produksi tanaman padi lima tahun terakhir di Kabupaten Grobogan mengalami naik turun
setiap tahunnya. Jika produksi padi menurun maka akan meyebabkan persediaan pangan menjadi
tidak stabil. Pemerintah Daerah Kabupaten Grobogan harus mengupayakan kebijakan agar
produksi padi dapat selalu terpenuhi. Salah satu caranya adalah melihat perkiraan produksi padi
pada tahun berikutnya dapat dilakukan dengan melakukan peramalan. Namun peramalan produksi
padi yang dilakukan oleh Badan Pusat Statistik (BPS) Kabupaten Grobogan saat ini masih
menghasilkan tingkat error yang masih tinggi dan hasil ramalan masih jauh berbeda dengan data
aktual. Dengan demikian, tantangan utama dalam peramalan produksi padi adalah untuk
menghasilkan metode realistis yang harus mampu menangani data deret waktu yang
kompleks dan menghasilkan perkiraan dengan kesalahan yang sangat kecil. As such, main
challenge in rice production forecasting is to generate realistic method that must be capable
for handling complex time series data and generating forecasting with almost tiny error.

Fuzzy time series telah dilaksanakan untuk prediksi data di berbagai sektor, seperti
pendidikan, keuangan-ekonomi, energi, kecelakaan lalu lintas, lainnya. Selain itu, banyak
model yang diusulkan telah disajikan untuk meningkatkan akurasi peramalan. Namun,
penyesuaian interval-panjang dan prosedur perkiraan out-sample masih menjadi masalah
dalam peramalan time series fuzzy, di mana kedua masalah belum diselidiki secara jelas

1
dalam studi sebelumnya. Fuzzy time series has been implemented for data prediction in the
various sectors,
such as education, finance-economic, energy, traffic accident, others. Moreover, many
proposed models have been presented to improve the forecasting accuracy. However,
the interval-length adjustment and the out-sample forecast procedure are still issues in
fuzzy time series forecasting, where both issues are yet clearly investigated in the previous
studies.

Metode Average Based Fuzzy Time Series digunakan untuk meramalkan produksi
padi di Kabupaten Grobogan. Data yang digunakan pada peramalan ini yaitu data produksi padi
setiap tahun Kabupaten Grobogan dari tahun 1992 sampai dengan tahun 2016.
The results of paddy production in the last five years in Grobogan Regency have experienced
ups and downs every year. If paddyproduction decreases, it will cause food supply to become
unstable. The Regional Government of Grobogan Regency must seek policies so that rice
production can always be fulfilled. One way is to see estimates of paddy production in the
following year can be done by forecasting. However, the forecasting of paddy production
carried out by the Central Statistics Agency (BPS) of Grobogan Regency currently still
produces a high level of error. As such, main challenge in rice production forecasting is to
generate realistic method that must be capable for handling complex time series data and
generating forecasting with almost tiny error. Fuzzy time series has been implemented for
data prediction in the various sectors, such as education, rainfall distribution, others.
Moreover, many proposed models have been presented to improve the forecasting accuracy.
However, the interval-length adjustment is still issues in fuzzy time series forecasting.
Therefore, this study explores the application of the average based fuzzy time series method
to obtain more accurate paddy production preditions. Data for the study were collected from
the Central Statistics Agency (BPS) Grobogan Regency. The data were analysed and
validated using the Mean Absolute Percentage Error (MAPE). To illustrate the forecasting
process and the effectiveness of the proposed model, two numerical datasets of paddy
production data every year and enrolment of students of Alabama University are examined.
The examined results show that the proposed model gets lower forecasting errors and has a
higher forecasting accuracy than those of other existing models.

2
 Dari pengujian menggunakan 25 data time series menghasilkan peramalan produksi
padi pada tahun 2017 sebesar 830 ribu ton dengan MAPE sebesar 5.44497% dan tingkat akurasi
peramalan 94.555%. Pada nilai MAPE nilai errornya <10%, sehingga dapat dikatakan metode
Average Based Fuzzy Time Series baik digunakan untuk peramalan khususnya peramalan produksi
padi di Kabupaten Grobogan karena mampu menghasilkan peramalan dengan tingkat akurasi yang
sangat tinggi.

Untuk mengilustrasikan proses peramalan keefektifan model yang diusulkan, dua dataset
numerik produksi padi rata-rata Vietnam dan pendaftaran mahasiswa Universitas Alabama
diperiksa. Hasil yang diperiksa menunjukkan bahwa model yang diusulkan memiliki
kesalahan peramalan yang lebih rendah dibandingkan dengan model lain yang ada. To
illustrate the forecasting process and the effectiveness of the proposed model, two numerical
datasets of average rice production of Viet Nam and enrolment of students of Alabama
University are examined. The examined results show that the proposed model gets lower
forecasting errors than those of other existing models

INTRODUCTION
Grobogan Regency is a rice production center in Central Java Province. The agricultural
sector has become one of the economic supports for the people of Grobogan Regency. The
availability of rice in Grobogan Regency continues to be considered because it is a source of food.
The development of rice production in the last five years in Grobogan Regency experienced a ups
and downs every year (Central Statistics Agency (BPS) of Grobogan Regency,2017). If rice
production decreases, food supply becomes unstable. Need to forecast rice production.
. Various techniques have been developed for time series forecasting in last few decade,
but the classical time series technique can not deal with the forecasting issues in which the data
of time series are represented by fuzzy sets (Chen & Hsu, 2004). The fuzzy time series forecasting
models based on fuzzy set theory (Zadeh, 1965). Fuzzy time series have been widely apllied to

3
diverse fields such as enrolments of university ol Alabama forecasting (Song & Chissom, 1993b),
crop productions prediction (Singh, 2007) and rainfall distribution (Othman & Azahari, 2016). The
fuzzy time series and the corresponding forecast model was introduced by Song and Chissom in
1993. Then, in 1994, they proposed a new fuzzy time series and compared three different
defuzzification models (Song & Chissorn, 1994). Chen (1996) resolved the problem of large
computional requirements of Song and Chissom method of computing fuzzy relations using max-
min composition by replacing it with the simplified arithmatic operations and applied the method
on the student enrollments of University of Alabama. The arithmetic operations have a robust
specification and are superior to those applied in Song and Chissom model. Following these
definitions, fuzzy time series models have been proposed for various applications, such as
enrollment (Chen, 1996) (Huarng, 2001) (Song & Chissom, 1993a) (Song & Chissom, 1993b)
(Song & Chissorn, 1994) and temperature forecasting (Chen, 2000).
Then Chen (2000) implemented the Fuzzy Time Series method to forecast the
temperature to be carried out after the previous temperature data. After that, the fuzzy time
series has been widely studied to improve forecasting accuracy. Furthermore Huarng (2001)
presented an effective approaches that can adjust lenghts intervals to get better forecasting
accuracy. That study may result in different fuzzy relationships, and in turn different
forecasting results. And the effective lengths of intervals do improve forecasting results in
various empirical analyses. Kumar & Kumar (2012) had apllied the fuzzy time series in their
studeis to forecast rice production. Othman & Azahari (2016) presented a new model to
forecast ranfall distribution in Malaysia by using 168 rainfall distribution data. Abhisekh &
Kumar (2017) apllied a computational method for rice production forecasting based on high-
order fuzzy time series. In the present work to proposed average-based fuzzy time series
models was implemented on the historical paddy production data in Grobogan Regency to
improve fuzzy time series forecasting.

In this ubsequent sections, the basic concept of fuzzy time series and average based
fuzzy time series are explained in detail. Then, the methodology of the average based fuzzy
time series model of paddy productions forecasting is discussed. Next, the result and
discussion of this study is highlighted followes by the conclusion that is presented in the last
section.

4
Basic Concept of Fuzzy Time Series
Song and Chissom first proposed the definitions of fuzzy time series in 1993 (Song & Chissom,
1993b). The concepts of fuzzy time series are described as follows.
Let U be the universe of discourse, where U = {u1, u2, . . . , un}. A fuzzy set Ai of U is defined
by Ai = fAi (u1)/u1 + f Ai(u2)/u2 + . . . fAi(un)/un, where fAis the membership function of the
fuzzy set Ai, ; f
.

5
Different fuzzy time-series models have been proposed by following Song and Chissom’s
definition of fuzzy time-series. For example, there have been Song and Chissom’s time-invariant
model (Song & Chissom, 1993a), Chen’s model (Chen, 1996), etc.
It is common for these models to include the following steps: (1) define the universe of discourse
andthe intervals for the observations; (2) partition the universe based on the intervals; (3) define
the fuzzy sets
for the observations; (4) fuzzify the observations; (5) establish the fuzzy logic relationship and
fuzzy logic relationship group; (6) perform the forecast; and (7) defuzzify the forecasting results.
Average Based Fuzzy Time Series Model
In previous models, all the lengths of the intervals were determined at Step 1 of the forecasting
process.The interval length affects the formulation of fuzzy relationships, and the fuzzy
relationships affect the forecasting results. In addition to capture proper fuzzy relationships, the
determination of proper interval lengths is also critical in fuzzy time-series forecasting. A key point
in choosing effective lengths of intervals is that they should not be too large or small (Huarng,
2001). While the length of intervals is too large, there will be no fluctuations in the fuzzy time
series. On the other hand, when the length is too small, the meaning of fuzzy time series will be
diminished. Thus, the lengths should reflect at least half the fluctuations in the time series (Huarng,
2001).
However, the fluctuations in fuzzy time series can be represented by the absolute value of
the first differences of any two consecutive data (Huarng, 2001). Hence, the average-based length
is proposed.
Average Based Lenght
In this section, we propose an average-based fuzzy time-series model, which can be used
to adjust the lengths of the intervals determined during the early stages of forecasting, when the
fuzzy elationships are formulated.
Following the discussion in Section 3.1, the algorithm for average-based lengths is
demonstrated as follows (Huarng, 2001):
(1). Calculate all the absolute differences between Ai+1 and Ai (i = 1, . . . , n − 1), as the first
differences and the average of the first differences.
(2). Take one half the average (in step 1) as the length.

6
(3). According to the length (in step 2), determine the base for the length of intervals by following
Tab.
Table 1. base mapping table
Range Base
0.1-1.0 0.1
1.1-10 1
11-100 10
101-1000 100
(4). Round the length according to the determined base as the length of intervals.
In order to show how to calculate the average-based length, an example is given. Suppose we have
the following time series data: 30,50,80,120,110, and 70. The algorithm for average-based length
is implemented step by step below:
(1). The first differences are 20, 30, 40, 10, 40
The average of the first differences is 28.
(2). Take half of the average as the length, which is 14.
(3). According to the length (in step 2), the base for length of intervals is determined as 10 by Tab.
1.
(4). Round the length 14 by the base 10, which is 10. So 10 is chosen as the length of interval.

METHODOLOGY
Forecasting with average-based model
Produksi padi digunakan untuk mendemonstrasikan peramalan mengguna kan fuzzy time series
model. Data produksi padi yang digunakan adalah data produksi padi setiap tahun dari tahun
1992 sampai dengan tahun 2016. Metode peramalan produksi padi terdiri dari 3 langkah. Dapat
dilihat pada gambar Flowchart Peramalan Produksi Padi

7
 Flowchart Peramalan Produksi Padi

Phase 1 : Data Collection and Processing

The historical data of produksi padi setiap tahun Kabupaten Grobogan dari tahun 1992 sampai
dengan tahun 2016 dalam ribu ton yang diperoleh dari Dinas Pertanian TPH dan BPS Kabupaten
Grobogan. Pola data dapat dilihat pada gambar

8
 Data Produksi Padi Aktual
900
800
700
600
500
400
300
200
100
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25

Grafik data aktual produksi padi

Tabel 4.1 Data Aktual Produksi Padi di Kabupaten Grobogan dari tahun 1992 sampai
tahun 2016

Produksi Produksi Produksi

Padi Padi Padi
No Tahun No Tahun No Tahun
(ribu (ribu (ribu
ton) ton) ton)
1 1992 446.177 9 2000 450.273 17 2008 627.766
2 1993 436.557 10 2001 508.304 18 2009 659.314
3 1994 453.234 11 2002 556.869 19 2010 663.7756
4 1995 452.044 12 2003 554.873 20 2011 595.5382
5 1996 482.187 13 2004 540.078 21 2012 608.7498
6 1997 494.13 14 2005 512.341 22 2013 629.0769
7 1998 524.011 15 2006 582.515 23 2014 554.5863
8 1999 380.51 16 2007 561.08 24 2015 786.0884
25 2016 827.5095

Phase 2 : Forecasting Produksi Padi Based On Average based Fuzzy Time Series

Step 1. Defining the universe of discourse and intervals.

The universe of discourse for observations, U , is defined as [Dmin − D1, Dmax − D2], where
Dmin and Dmax is the minimum and maximum of known historical data, D1, D2 are two proper

9
positive numbers. According to the SCSI for 2004, we can see that Dmin = 1260.316 and Dmax =
1777.516. Thus, the universe of discourse is defined as U = [1260, 1780].
1. Mengumpulkan data, menentukan nilai himpunan semesta dan membagi himpunan semesta

menjadi beberapa interval

Pada penelitian ini, himpunan semesta dibagi menjadi beberapa interval dengan

panjang interval ditentukan berdasarkan rata-rata (average based lenght). Dari 25 data

produksi padi di Kabupaten Grobogan pada Tabel 4.1, diperoleh data terkecil atau

𝑋(𝑡) 𝑚𝑖𝑛 = 380.51, data terbesar atau 𝑋(𝑡) 𝑚𝑎𝑥 = 827.509, dan sembarang data bilangan

positif 𝑝1 = 0.5 dan 𝑝2 = 12.4905. Sehingga diperoleh himpunan semesta berikut ini.

𝑈 = [380 , 840]

Average-based length is calculated:

(1). The first differences are calculated and the average of the first difference is 31.94138.
(2). Take half the average as the length, which is 15.97069.
(3). Given the length chosen in step 2, the base for length of intervals is determined as 10 by
following Tab. 1.
(4). Round the length by base 10, which is 20. 20 is chosen as the length.
Now average-based length is applied to Chen’s model.
Then U can be partitioned into equal-length intervals u1, . . . , u26, and the midpoints of these
intervals are m1, . . . , m26, respectively, where u1= [1260, 1280], . . . , u2= [1760, 1780].

2. Dengan menggunakan aturan penentuan panjang interval berbasis rata-rata (average based

lenght) berdasarkan Gambar 3.6, yaitu:

1) Menghitung selisih mutlak (absolute) setiap data produksi padi antara tahun n dan tahun

n+1, maka dapat diketahui bahwa rata-rata selisih mutlak data adalah 46.139015.

2) Menentukan setengah rata-rata langkah pertama sebagai panjang interval, yaitu

23.069507.

10
 3) Sesuai Tabel 2.1, maka 23.069507 termasuk dalam kategori interval basis 10.

4) Membulatkan nilai 23.069507 dengan menggunakan basis 10, maka diperoleh angka 20

sebagai panjang interval.

Dari hasil perhitungan di atas, panjang interval (range) sebesar 20, maka diperoleh interval
sebanyak 23. Interval-interval yang terbentuk adalah sebagai berikut

𝑢1 = [380,400), 𝑢2 = [400,420), 𝑢3 = [420,440),

𝑢4 = [440,460), 𝑢5 = [460,480), 𝑢6 = [480,500],

𝑢7 = [500,520), 𝑢8 = [520,540), 𝑢9 = [540,560),

𝑢10 = 560,580), 𝑢11 = [580,600), 𝑢12 = [600,620),

𝑢13 = [620,640), 𝑢14 = [640,660), 𝑢15 = [660,680),

𝑢16 = [680,700), 𝑢17 = [700,720), 𝑢18 = [720,740),

𝑢19 = [740,760), 𝑢20 = [760,780), 𝑢21 = [780,800),

𝑢22 = [800,820), 𝑢23 = [820,840].

Step 2. Defining fuzzy sets for observations.

Each linguistic observation, Ai
, can be defined by the interval u1, . . . , u26, as follows:
A1= 1/u1+ 0.5/u2+ 0/u3+ 0/u4 + . . . + 0/u26;
A2= 0.5/u1+ 1/u2+ 0.5/u3+ 0/u4 + . . . + 0/u26;
A2= 0 + 0.5/u2+ 1 + 0.5/u4 + . . . + 0/u26;
...
A25= 0/u1+ 0/u2+ 0/u3+ 0/u4 + . . . + 0.5/u24+ 1/u25+ 0.5/u26;
A26= 0/u1+ 0/u2+ 0/u3+ 0/u4 + . . . + 0/u24+ 0.5/u25+ 1/u26;

11
 Interval dan himpunan fuzzy bernilai sama, karena interval sesuai dengan variabel

linguistik. Himpunan-himpunan fuzzy didefinisikan menurut fungsi keanggotaan fuzzy.

1 0,5
+ , 𝑖=1
𝑢1 𝑢2
0,5 1 0,5
𝐴𝑖 = + + , 2≤𝑖 ≤𝑘−1
𝑢𝑖−1 𝑢𝑖 𝑢𝑖+1
0,5 1
+ , 𝑖=𝑘
{ 𝑢𝑘−1 𝑢𝑘

𝑓(𝑢𝑖 )
, 𝑓(𝑢𝑖 ) merupakan derajat keanggotaan interval 𝑢𝑖 dalam himpunan 𝐴𝑖 . Pada
𝑢𝑖

𝑖 = 1 diperoleh himpunan fuzzy 𝐴1 (himpunan fuzzy produksi padi yang paling rendah). Pada

𝑖 = 𝑘, diperoleh himpunan fuzzy 𝐴𝑘 (himpunan fuzzy produksi padi yang paling tinggi).

Semakin besar nilai 𝑖, himpunan fuzzy produksi padi akan bergerak dari yang paling rendah

menjadi himpunan fuzzy produksi padi yang paling tinggi. Pada penelitian ini, fungsi

keanggotaan untuk setiap himpunan fuzzy adalah sebagai berikut.

1) Fungsi Keanggotaan himpunan fuzzy 𝐴1

1 0,5 0 0 0 0 0 0 0 0
𝐴1 = + + + + + + + + + +
𝑢1 𝑢2 𝑢3 𝑢4 𝑢5 𝑢6 𝑢7 𝑢8 𝑢9 𝑢10

0 0 0 0 0 0 0 0 0
+ + + + + + + + +
𝑢11 𝑢12 𝑢13 𝑢14 𝑢15 𝑢16 𝑢17 𝑢18 𝑢19

0 0 0 0
+ + +
𝑢20 𝑢21 𝑢22 𝑢23

Fungsi Keanggotaan Fuzzy 𝐴1 dapat dilihat pada Gambar 4.1.

12
 Gambar 4.1 Fungsi Keanggotaan Fuzzy 𝐴1

Dengan kata lain, fungsi keanggotaan fuzzy untuk himpunan fuzzy 𝐴1 dengan

derajat keanggotaan 1 untuk interval 𝑢1 , derajat keanggotaan 0,5 untuk interval 𝑢2 dan

derajat keanggotaan 0 untuk interval 𝑢3 sampai interval 𝑢23 . Sehingga diperoleh

himpunan fuzzy 𝐴1 = [380,400).

Step 3. Fuzzifying observations.

Each SCSI can be fuzzified into a fuzzy set. To facilitate the explanations that follow, some SCSI
and their corresponding fuzzy set Ai are listed in Tab. 2.
Step 4. Establishing FLRs and FLRGs.
Data Himpunan Data Himpunan
No No
Aktual Fuzzy Aktual Fuzzy
1 446.177 𝐴4 14 512.341 𝐴7
2 436.557 𝐴3 15 582.515 𝐴11
3 453.234 𝐴4 16 561.08 𝐴10
4 452.044 𝐴4 17 627.766 𝐴13
5 482.187 𝐴6 18 659.314 𝐴14
6 494.13 𝐴6 19 663.7756 𝐴15
7 524.011 𝐴8 20 595.5382 𝐴11
8 380.51 𝐴1 21 608.7498 𝐴12
9 450.273 𝐴4 22 629.0769 𝐴13
10 508.304 𝐴7 23 554.5863 𝐴9
11 556.869 𝐴9 24 786.0884 𝐴21
12 554.873 𝐴9 25 827.5095 𝐴23
13 540.078 𝐴9

13
Berdasarkan Tabel 4.3, data produksi padi ke-1 yaitu 446.177 berada pada interval [440,460],
sedangkan interval [440,460] sesuai dengan himpunan fuzzy 𝐴4 . Sehingga data 446.177 dapat
diubah menjadi himpunan fuzzy 𝐴4 . Analog untuk data kedua sampai data ke 25

Based on the fuzzy set in step 3, the FLRs are established, as in Tab. 3. The FLRs can be rearranged
to FLRGs, as in Tab. 4.
Step 5. Forecasting
FLR

No Time Series FLR No Time Series FLR

1 1992 – 1993 𝐴4 → 𝐴3 13 2004 – 2005 𝐴9 → 𝐴7

2 1993 – 1994 𝐴3 → 𝐴4 14 2005 – 2006 𝐴7 → 𝐴11
3 1994 – 1995 𝐴4 → 𝐴4 15 2006 – 2007 𝐴11 → 𝐴10
4 1995 – 1996 𝐴4 → 𝐴6 16 2007 – 2008 𝐴10 → 𝐴13
5 1996 – 1997 𝐴6 → 𝐴6 17 2008 – 2009 𝐴13 → 𝐴14
6 1997 – 1998 𝐴6 → 𝐴8 18 2009 – 2010 𝐴14 → 𝐴15
7 1998 – 1999 𝐴8 → 𝐴1 19 2010 – 2011 𝐴15 → 𝐴11
8 1999 – 2000 𝐴1 → 𝐴4 20 2011 – 2012 𝐴11 → 𝐴12
9 2000 – 2001 𝐴4 → 𝐴7 21 2012 – 2013 𝐴12 → 𝐴13
10 2001 – 2002 𝐴7 → 𝐴9 22 2013 – 2014 𝐴13 → 𝐴9
11 2002 – 2003 𝐴9 → 𝐴9 23 2014 – 2015 𝐴9 → 𝐴21
12 2003 – 2004 𝐴9 → 𝐴9 24 2015 – 2016 𝐴21 → 𝐴23

FLRG
No Sisi Kiri Sisi Kanan
(Kondisi Saat Ini) (Kondisi Selanjutnya)
1 𝐴1 𝐴4
2 𝐴3 𝐴4
3 𝐴4 𝐴3 , 𝐴4 , 𝐴6, , 𝐴7
4 𝐴6 𝐴6 , 𝐴8
5 𝐴7 𝐴9 , 𝐴11
6 𝐴8 𝐴1
7 𝐴9 𝐴7 , 𝐴9 , 𝐴21
8 𝐴10 𝐴13

14
 9 𝐴11 𝐴10 , 𝐴12
10 𝐴12 𝐴13
11 𝐴13 𝐴9 , 𝐴14
12 𝐴14 𝐴15
13 𝐴15 𝐴11
14 𝐴21 𝐴23

Peramalan 2017
Data Data Hasil
Keanggotaan
No Tahun Aktual Peramalan |𝑨𝒕 -𝑭𝒊 | |𝑨𝒕 -𝑭𝒊 |/𝑨𝒕
Fuzzy
(𝑨𝒕 ) (𝑭𝒊 )
1 1992 446.177 𝐴4 -
2 1993 436.557 𝐴3 470.000 33.443 0.076606262
3 1994 453.234 𝐴4 450.000 3.234 0.007135387
4 1995 452.044 𝐴4 470.000 17.956 0.039721797
5 1996 482.187 𝐴6 470.000 12.187 0.025274427
6 1997 494.13 𝐴6 510.000 15.87 0.032117054
7 1998 524.011 𝐴8 510.000 14.011 0.026737988
8 1999 380.51 𝐴1 390.000 9.49 0.024940212
9 2000 450.273 𝐴4 450.000 0.273 0.000606299
10 2001 508.304 𝐴7 470.000 38.304 0.07535648
11 2002 556.869 𝐴9 570.000 13.131 0.023580052
12 2003 554.873 𝐴9 616.667 61.7937 0.111365484
13 2004 540.078 𝐴9 616.667 76.5887 0.141810442
14 2005 512.341 𝐴7 616.667 104.3257 0.203625515
15 2006 582.515 𝐴11 570.000 12.515 0.021484425
16 2007 561.08 𝐴10 590.000 28.92 0.051543452
17 2008 627.766 𝐴13 630.000 2.234 0.003558651
18 2009 659.314 𝐴14 600.000 59.314 0.089963204
19 2010 663.7756 𝐴15 670.000 6.2244 0.009377265
20 2011 595.5382 𝐴11 590.000 5.5382316 0.00929954
21 2012 608.7498 𝐴12 590.000 18.749844 0.030800573
22 2013 629.0769 𝐴13 630.000 0.923121207 0.001467422
23 2014 554.5863 𝐴9 600.000 45.413684 0.081887495
24 2015 786.0884 𝐴21 616.667 169.4213533 0.215524569
25 2016 827.5095 𝐴23 830.000 2.4905063 0.003009641
Jumlah 752.3512404 1.306793635

Forecasting is conducted by the following rules:

Rule 1: If the current fuzzy set is Ai , and the fuzzy logical relationship group of Ai is empty, i.e.,
Ai →,the forecast is mi , the midpoint of ui.
Forecasting = mi
15
Rule 2: If the current fuzzy set is Ai , and the fuzzy logical relationship group of Ai is one-to-
one, i.e.,Ai → Aj, the forecast is mj , the midpoint of uj .
Forecasting = mj
Rule 3: If the current fuzzy set is Ai , and the fuzzy logical relationship group of Ai is one-to-
many,i.e., Ai → Aj1, Aj2, . . . , Ajn, the forecast is equal to the average of mj1, mj2, . . . , mjn,
the midpoints of uj1, uj2, . . . , ujn, respectively.
Forecasting =Pni=1mjin

3. Melakukan defuzzifikasi hasil peramalan dengan aturan sebagai berikut. Misalkan 𝐹(𝑡)

adalah data yang akan diramalkan dimana 𝐹(𝑡 − 1) = 𝐴𝑖 , maka:

a. Jika hanya terdapat satu Fuzzy Logical Relationship Group (FLRG) dari 𝐴𝑖 yaitu 𝐴𝑖 →

𝐴𝑗 , maka 𝐹(𝑡) = 𝐴𝑗 dimana defuzzifikasinya adalah nilai tengah dari interval yang

memiliki nilai keanggotaan maksimum pada 𝐴𝑗 .

Misalkan 𝐹(𝑡 − 1) = 𝐴1 dan Fuzzy Logical Relationship Group (FLRG) dari 𝐴1

yaitu 𝐴1 → 𝐴4 , maka defuzzifikasinya adalah nilai tengah dari interval yang memiliki

nilai keanggotaan maksimum pada 𝐴4 yaitu interval [440,460). Sehingga hasil

defuzzifikasinya adalah 450. Analog untuk himpunan fuzzy lainnya.

b. Jika 𝐴𝑖 tidak memiliki relationship yaitu 𝐴𝑖 → ∅ maka defuzzifikasi 𝑋(𝑡) diperoleh dari

nilai tengah interval yang memiliki nilai keanggotaan maksimum pada 𝐴𝑖 .

c. Jika terdapat lebih dari satu Fuzzy Logical Relationship Group dari 𝐴𝑖 yaitu 𝐴𝑖 →

𝐴𝑘1 , 𝐴𝑘2 , … maka defuzzifikasi 𝐹(𝑡) diperoleh dari rata-rata nilai tengah dari masing-

masing interval yang memiliki nilai keanggotaan maksimum pada masing-masing

𝐴𝑘1 , 𝐴𝑘2 , ….

| 𝐷𝑎𝑡𝑎 𝐴𝑘𝑡𝑢𝑎𝑙−𝐷𝑎𝑡𝑎 𝑅𝑎𝑚𝑎𝑙|

∑𝑛
𝑘=1 𝐷𝑎𝑡𝑎 𝐴𝑘𝑡𝑢𝑎𝑙
𝑀𝐴𝑃𝐸 = × 100
𝑛

16
Defuzzifikasi

Misalkan 𝐹(𝑡 − 1) = 𝐴1 dan Fuzzy Logical Relationship Group (FLRG) dari 𝐴6

yaitu 𝐴6 → 𝐴6 , 𝐴8 , maka defuzzifikasinya adalah rata-rata nilai tengah dari interval yang

memiliki nilai keanggotaan maksimum pada 𝐴6 dan 𝐴8 yaitu interval [480,500) dan

interval [520,540). Sehingga hasil defuzzifikasinya adalah

490 + 530
= 510
2

1.306793635
𝑀𝐴𝑃𝐸 = × 100 = 5.44497348%
24

Phase 3 : Evaluate the forecasting models

17
 Pada prinsipnya, pengawasan peramalan dilakukan dengan membandingkan hasil

peramalan dengan kenyataan yang terjadi. Penggunaan teknik peramalan yang menghasilkan

penyimpangan terkecil adalah teknik peramalan yang sesuai untuk digunakan. Kriteria yang

digunakan untuk mengukur ketepatan metode peramalan adalah Mean Absolute Percentage Error

(MAPE). MAPE merupakan salah satu metode yang dapat digunakan untuk menghitung nilai

kesalahan peramalan dengan cara menghitung kesalahan mutlak (absolute) pada setiap periode

kemudian dibagi dengan nilai observasi yang dilakukan pada periode tersebut (Pramita Lucianna,

2017:31).

Rumus untuk menghitung MAPE adalah sebagai berikut.

| 𝐷𝑎𝑡𝑎 𝐴𝑘𝑡𝑢𝑎𝑙−𝐷𝑎𝑡𝑎 𝑅𝑎𝑚𝑎𝑙|
∑𝑛
𝑘=1 𝐷𝑎𝑡𝑎 𝐴𝑘𝑡𝑢𝑎𝑙
𝑀𝐴𝑃𝐸 = × 100 (2.9)
𝑛

Keterangan:

- Data Aktual merupakan data asli atau data mentah yang belum diolah

- Data Ramal merupakan data hasil peramalan yang telah dilakukan

- n merupakan banyaknya periode ramalan

MAPE merupakan metode yang digunakan dalam menghitung kesalahan peramalan dengan

menetapkan error mutlak sebagai presentase dari rata-rata error mutlak untuk sejumlah periode

data aktual. Berikut merupakan kriteria MAPE (Rahmadiani dan Anggraeni, 2012:3).

RESULT AND DISCUSSION

18