its undergraduate 9832 presentation

35
  PENERAPAN METODE EXTREME LEARNING MACHINE (EL M) UNTUK PERAMA L A N PERMINT A A N Pen usun Tu as Akhir : Irwin Dwi A. (NRP : 5204.100.077) Dosen Pembimbing : Wiw iek Ang gr aeni S.si.,M.Kom   . ., 29 Januari 2010

Upload: fadharzlite

Post on 20-Jul-2015

57 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: ITS Undergraduate 9832 Presentation

5/17/2018 ITS Undergraduate 9832 Presentation - slidepdf.com

http://slidepdf.com/reader/full/its-undergraduate-9832-presentation 1/35

PENERAPAN MET DE EXTREME LEARNIN MA HINE 

(ELM) UNTUK PERAMALAN PERMINTAAN

Pen usun Tu as Akhir :  

Irwin Dwi A.(NRP : 5204.100.077)

Dosen Pembimbing :  

Wiwiek Anggraeni S.si.,M.Kom  . .,

29 Januari 2010

Page 2: ITS Undergraduate 9832 Presentation

5/17/2018 ITS Undergraduate 9832 Presentation - slidepdf.com

http://slidepdf.com/reader/full/its-undergraduate-9832-presentation 2/35

 

• Memperkirakan permintaan konsumen di masa datang selalu

. : a ar e a ang :.

menj adi tantangan bagi pelaku usaha dan indust ri.

• Peramalan an akurat dan efekt if da at membantu en ambilkeputusan dalam perusahaan menentukan j umlah barang yangakan diproduksi, bahan baku yang dibutuhkan sertamenentukan harga terhadap barang j adi.

• Jaringan syaraf t i ruan merupakan salah satu metode yangban ak dia likasikan dalam eramalan khususn a sales 

forecast ing .

Page 3: ITS Undergraduate 9832 Presentation

5/17/2018 ITS Undergraduate 9832 Presentation - slidepdf.com

http://slidepdf.com/reader/full/its-undergraduate-9832-presentation 3/35

 

• JST mempunyai beberapa kelebihan pada kont rol area,

. : a ar e a ang :.

prediksi dan pengenalan pola (Sun et al , 2008).

• ELM meru akan metode embela aran baru dari JSTfeedforward.

(Sun et al,2008), serta mempunyai t ingkat akurasi yang lebihbaik dibandingkan dengan metode konvensional .

Page 4: ITS Undergraduate 9832 Presentation

5/17/2018 ITS Undergraduate 9832 Presentation - slidepdf.com

http://slidepdf.com/reader/full/its-undergraduate-9832-presentation 4/35

 

Tuj uan dari penyusunan tugas akhir ini adalah

.: u uan ugas r:.

• Menerapkan metode Ext reme Learning Machine  (ELM) untukmeramalkan j umlah permintaan konsumen dengan data yangbersifat runtut waktu (time series).

Page 5: ITS Undergraduate 9832 Presentation

5/17/2018 ITS Undergraduate 9832 Presentation - slidepdf.com

http://slidepdf.com/reader/full/its-undergraduate-9832-presentation 5/35

 

Permasalahan an dian kat di dalam tu as akhir ini adalah

. : ermasa a an:.

• Bagaimana meramalkan j umlah permintaan konsumen denganmetode ext reme learning machine.

• Menghasilkan ramalan dengan t ingkat accuracy t inggi

Page 6: ITS Undergraduate 9832 Presentation

5/17/2018 ITS Undergraduate 9832 Presentation - slidepdf.com

http://slidepdf.com/reader/full/its-undergraduate-9832-presentation 6/35

 

• Data yang digunakan adalah data penj ualan harian mil ik “ Cak

. : a asan ermasa a an:.

Cuk Shop” Surabaya periode 2008-2009.

• Peramalan dibatasi ada dua roduk aitu kaos dan in

• Pengembangan aplikasi menggunakan ruang lingkup.

Page 7: ITS Undergraduate 9832 Presentation

5/17/2018 ITS Undergraduate 9832 Presentation - slidepdf.com

http://slidepdf.com/reader/full/its-undergraduate-9832-presentation 7/35

 . : x reme earn ng ac ne :.

• Ext reme learning machine (ELM) merupakan skemapembelaj aran baru dari j aringan syaraf t iruan.

• ELM merupakan j aringan syaraf t i ruan feedforward dengansingle hidden layer atau biasa disebut dengan Single Hiddel LayerFeedforward neural Networks (SLFNs).

• Parameter-parameter sepert i input weight dan hidden bias

dipil ih secara random.

Page 8: ITS Undergraduate 9832 Presentation

5/17/2018 ITS Undergraduate 9832 Presentation - slidepdf.com

http://slidepdf.com/reader/full/its-undergraduate-9832-presentation 8/35

 . : o e a ema s :.

Untuk j umlah sample yang berbeda dengan j umlah hidden),( ti Xi

nodes sebanyak N  dan activation function g(x)

 ji j

 N 

i

i

 N 

i

ob X W ig xjigi =+•= ∑∑==

).()(11

 β  β 

Dimana :

T wwww ...= Vektor weight yang menghubungkan t h hidden node sin an npu no es . npu we g

imiii ),...,,( 21 β  β  β  β = Vektor weight yang menghubungkan t h hidden nodes

dengan output (output weight)i

ibThreshold dari th hidden nodes

 ji xw

i

Hasil perkalian antara bobot (weight) dengan input

Page 9: ITS Undergraduate 9832 Presentation

5/17/2018 ITS Undergraduate 9832 Presentation - slidepdf.com

http://slidepdf.com/reader/full/its-undergraduate-9832-presentation 9/35

 . : o e a ema s :.

Dengan asumsi t ingkat error 0 makaot o N 

 j j =∑ − j j t o =

 ji j

 N 

i

i t b X W ig =+•∑=

).(1

 β 

 j=1

Rumus tersebut dapat ditulis sederhana

T  H  =

Dari rumus tersebut, Output weight dapat dihitung dengan rumus

T  H  Τ= β 

Page 10: ITS Undergraduate 9832 Presentation

5/17/2018 ITS Undergraduate 9832 Presentation - slidepdf.com

http://slidepdf.com/reader/full/its-undergraduate-9832-presentation 10/35

 . : e o o og :.

Langkah –langkah peramalan dengan metode Ext remeearn ng ac ne

Pembagian data menj adi data t raining dan test ing

 

Training ELM

Penentuan fun si akt ivasi dan umlah hidden neuron

Test ing : 20%dari total data 1})min{} /(max{})min{(2 −−×=− p p p p X  X  X  X  X 

X = nilai hasil normalisasi yang berkisar antara [-1,1]. 

Menghitung input weight , biass dan output weight

Denormalisasi output

.Min (Xp) = nilai minimum pada data set .

Max (Xp) = nilai maksimum pada data set .Input weight dan hidden biass ditentukan secara random

Τ

Test ing / peramalan dengan ELM 

= }min{})min{}(max{)1(5.0  p p p p X  X  X  X  X  +−×+×=X = nilai data setelah denormalisasi.X = data out ut sebelum denormalisasi.Min (Xp) = data minimum pada data set sebelum denormalisasi.

Max (Xp) = data maksimun pada data set sebelum normalisasi.

Page 11: ITS Undergraduate 9832 Presentation

5/17/2018 ITS Undergraduate 9832 Presentation - slidepdf.com

http://slidepdf.com/reader/full/its-undergraduate-9832-presentation 11/35

Page 12: ITS Undergraduate 9832 Presentation

5/17/2018 ITS Undergraduate 9832 Presentation - slidepdf.com

http://slidepdf.com/reader/full/its-undergraduate-9832-presentation 12/35

 . : mp emen as o a :.

“ ”

2008-2009 yang bersifat t ime series

Menggunakan tool Excel dan Mat lab 7.0.4

Skenario U i coba Menggunakan fungsi akt ivasi logsig, t ansig dan

purelin 

, ,

Page 13: ITS Undergraduate 9832 Presentation

5/17/2018 ITS Undergraduate 9832 Presentation - slidepdf.com

http://slidepdf.com/reader/full/its-undergraduate-9832-presentation 13/35

 . : mp emen as o a :.

 

Mean Absolut Precentage Error (MAPE)membandingkan prosentase kesalahan nilai rata-rata

a so ut antara n a perama an engan n a yangsebenarnya terj adi.

1 ar−net t rediksi Y Y 

Yprediksi = nilai prediksi

.1 arg=

=i et t Y n

Ytarget = nilai sesungguhnya yang terj adi.n = j umlah data yang diproses.

Page 14: ITS Undergraduate 9832 Presentation

5/17/2018 ITS Undergraduate 9832 Presentation - slidepdf.com

http://slidepdf.com/reader/full/its-undergraduate-9832-presentation 14/35

 . : mp emen as o a :.

Mean Square Error

Parameter ini merupakan rata-rata kesalahan yangmerupakan selisih dari nilai prediksi dengan data.

∑=−=

 N 

iii t  y N mse 1

2

)(

1

Keterangan :

 j umlah data= N 

data ouput (predicted sales)data penj ualan aktual

=i y=it 

Page 15: ITS Undergraduate 9832 Presentation

5/17/2018 ITS Undergraduate 9832 Presentation - slidepdf.com

http://slidepdf.com/reader/full/its-undergraduate-9832-presentation 15/35

 . : mp emen as o a :.

MSE dan MAPE hasil analisis dari ELM dibandingakndengan MSE dan MAPE dari metode Exponent ialSmoothing dan Moving Average.

Menggunakan metode Single Exponent ial Smoothing

dan Single Moing Average 2 periode.

Page 16: ITS Undergraduate 9832 Presentation

5/17/2018 ITS Undergraduate 9832 Presentation - slidepdf.com

http://slidepdf.com/reader/full/its-undergraduate-9832-presentation 16/35

 . : as o a:.

Pada produk kaos

Training ELM

Parameter bobot dan biass dengan t ingkat akurasi opt imaldidapatkan pada fungsi akt ivasi purelin j umlah hidden neuron 5

Tingkat AccuracyMAPE 0.04MSE 0.04

Page 17: ITS Undergraduate 9832 Presentation

5/17/2018 ITS Undergraduate 9832 Presentation - slidepdf.com

http://slidepdf.com/reader/full/its-undergraduate-9832-presentation 17/35

. : as o a:.

Peramalan dengan metode ELM

Tingkat Accuracy= .

MAPE = 0.0042

Page 18: ITS Undergraduate 9832 Presentation

5/17/2018 ITS Undergraduate 9832 Presentation - slidepdf.com

http://slidepdf.com/reader/full/its-undergraduate-9832-presentation 18/35

 . : as o a:.

Pada produk kaos

Peramalan dengan metode MA

Tingkat AccuracyMSE = 116.74MAPE = 19.19

Page 19: ITS Undergraduate 9832 Presentation

5/17/2018 ITS Undergraduate 9832 Presentation - slidepdf.com

http://slidepdf.com/reader/full/its-undergraduate-9832-presentation 19/35

 . : as o a:.

Peramalan den an metode ES

Tingakat Accuracy : MSE = 502.19 MAPE = 39.23

Page 20: ITS Undergraduate 9832 Presentation

5/17/2018 ITS Undergraduate 9832 Presentation - slidepdf.com

http://slidepdf.com/reader/full/its-undergraduate-9832-presentation 20/35

. : as o a:.

Perbandingan Metode ELM, ES dan MA

Page 21: ITS Undergraduate 9832 Presentation

5/17/2018 ITS Undergraduate 9832 Presentation - slidepdf.com

http://slidepdf.com/reader/full/its-undergraduate-9832-presentation 21/35

. : as o a:.

Perbandingan Nilai MSE dan MAPE Metode ELM,

ES dan MA

MA ES ELM

MSE 116.74 502.19 0.0481

MAPE 19.19 32.93 0.0042

Page 22: ITS Undergraduate 9832 Presentation

5/17/2018 ITS Undergraduate 9832 Presentation - slidepdf.com

http://slidepdf.com/reader/full/its-undergraduate-9832-presentation 22/35

. : as o a:.

Rata-rata t raining pada j umlah hidden neuron yang

berbeda

 

1 0.020866667

3 0.00525 0.010433333

7 0

Page 23: ITS Undergraduate 9832 Presentation

5/17/2018 ITS Undergraduate 9832 Presentation - slidepdf.com

http://slidepdf.com/reader/full/its-undergraduate-9832-presentation 23/35

 . : as o a:.

Pada produk pin

Training ELM

Parameter bobot dan biass dengan t ingkat akurasi opt imaldidapatkan pada fungsi akt ivasi purelin j umlah hidden neuron 5

Tingkat AccuracyMAPE 0.8480MSE 0.0021

Page 24: ITS Undergraduate 9832 Presentation

5/17/2018 ITS Undergraduate 9832 Presentation - slidepdf.com

http://slidepdf.com/reader/full/its-undergraduate-9832-presentation 24/35

. : as o a:.

Peramalan dengan metode ELM

Tingkat Accuracy= .

MAPE = 0.0095

Page 25: ITS Undergraduate 9832 Presentation

5/17/2018 ITS Undergraduate 9832 Presentation - slidepdf.com

http://slidepdf.com/reader/full/its-undergraduate-9832-presentation 25/35

. : as o a:.

Peramalan dengan metode MA

Tingkat AccuracyMSE 503.81MAPE 43.36

Page 26: ITS Undergraduate 9832 Presentation

5/17/2018 ITS Undergraduate 9832 Presentation - slidepdf.com

http://slidepdf.com/reader/full/its-undergraduate-9832-presentation 26/35

 . : as o a:.

Peramalan den an metode ES

Tingakat Accuracy : MSE = 55.45 MAPE = 111.39

Page 27: ITS Undergraduate 9832 Presentation

5/17/2018 ITS Undergraduate 9832 Presentation - slidepdf.com

http://slidepdf.com/reader/full/its-undergraduate-9832-presentation 27/35

. : as o a:.

Perbandingan Metode ELM, ES dan MA

Page 28: ITS Undergraduate 9832 Presentation

5/17/2018 ITS Undergraduate 9832 Presentation - slidepdf.com

http://slidepdf.com/reader/full/its-undergraduate-9832-presentation 28/35

. : as o a:.

Perbandingan Nilai MSE dan MAPE Metode ELM,

ES dan MA

MA E ELM

MSE 13.78 55.45 0.0023MAPE 55.43 111.39 0.0095

Page 29: ITS Undergraduate 9832 Presentation

5/17/2018 ITS Undergraduate 9832 Presentation - slidepdf.com

http://slidepdf.com/reader/full/its-undergraduate-9832-presentation 29/35

. : as o a:.

Rata-rata t raining pada j umlah hidden neuron yang

berbeda

Jumlah hidden neuron Trainin Time

1 0.0052

3 0

7 0

Page 30: ITS Undergraduate 9832 Presentation

5/17/2018 ITS Undergraduate 9832 Presentation - slidepdf.com

http://slidepdf.com/reader/full/its-undergraduate-9832-presentation 30/35

 . : es mpu an:.

Pada tugas akhir ini, ELM menghasilkan output opt imalpada fungsi akt ivasi purelin dengan j umlah hiddenneuron lima.

ELM menghasilkan output peramalan dengan t ingkat

kesalahan yang rendah daripada metode peramalankonvesional seperi Moving Average  dan Exponent ial Smoothing .

Page 31: ITS Undergraduate 9832 Presentation

5/17/2018 ITS Undergraduate 9832 Presentation - slidepdf.com

http://slidepdf.com/reader/full/its-undergraduate-9832-presentation 31/35

 . : es mpu an:.

Output dari ELM ditentukan oleh penentuaan parameterseperti fungsi akt ivasi atau fungsi t ransfer dan j umlah

hidden neuron.

Training time  atau learning speed  yang dibutuhkan

oleh ELM sangat singkat , yaitu rata- rata 0.00586667detik.

Page 32: ITS Undergraduate 9832 Presentation

5/17/2018 ITS Undergraduate 9832 Presentation - slidepdf.com

http://slidepdf.com/reader/full/its-undergraduate-9832-presentation 32/35

 . : aran:.

Implementasi metode ELM untuk masalah selain

peramalan seperti klasifikasi.

Menggunakan metode pembanding lain selain metode

Movin Avera e dan Ex onential Smoothin .

Menggunakan data input yang bersifat musiman atau

memiliki trend.

Page 33: ITS Undergraduate 9832 Presentation

5/17/2018 ITS Undergraduate 9832 Presentation - slidepdf.com

http://slidepdf.com/reader/full/its-undergraduate-9832-presentation 33/35

 . : a ar us a a:.

Huang, G.B., Zhu, Q.Y., dan Siew, C.K. 2004. Ext remeearn ng ac ne : ew earn ng c eme o e orwar

neural Networks. Proceeding of Internat ional JointConference on Neural Networks. Hungary, 25-29 Juli.

Makridakis, S., Wheelwright,S.C., dan McGee, V.E. 1999. Metode

dan Aplikasi Peramalan. Jakarta : Erlangga.

Mitchel, T.M. 1997. Machine learning. Singapura : McGraw‐Hill.

Qin‐Yu Zhu, A.K., Qin, P.N., Huang, G.B. 2005. Evolutionary Extreme

Learning Machine. Elsevier Pattern Recognition 38 (2005) 1759 ‐

1763.

Page 34: ITS Undergraduate 9832 Presentation

5/17/2018 ITS Undergraduate 9832 Presentation - slidepdf.com

http://slidepdf.com/reader/full/its-undergraduate-9832-presentation 34/35

 . : a ar us a a:.

Sun, Z.L., Choi, T.M., Au, K.F., dan Yu, Y. 2008. Sales Forecasting using

xtreme earn ng ac ne w t pp cat on n as on eta ng.

Elsevier Decision Support Systems 46 (2008) 411‐419.

Zhang, G., Pattuwo, B.E., dan Hu, M.Y. 1997. Forecasting with

Artificial Neural Networks : The State of the Art. Elsevier

International Journal of Forecasting 14 (1998) 35‐62.

Zhang, G.P. 2004. Neural Network Forecast ing in Bussiness.United States of America : Idea Group Publishing.

Page 35: ITS Undergraduate 9832 Presentation

5/17/2018 ITS Undergraduate 9832 Presentation - slidepdf.com

http://slidepdf.com/reader/full/its-undergraduate-9832-presentation 35/35

TERIMA KASIH