Articles

Found 24 Documents
Search

PREDIKSI TERJANGKITNYA PENYAKIT JANTUNG DENGAN METODE LEARNING VECTOR QUANTIZATION Hidayati, Nurul; Warsito, Budi
MEDIA STATISTIKA Vol 3, No 1 (2010): Media Statistika
Publisher : Jurusan Statistika FSM Undip

Show Abstract | Original Source | Check in Google Scholar | Full PDF (235.618 KB)

Abstract

Learning Vector Quantization (LVQ) is a method that train the competitives layer with supervised. The competitives layer will learn automatically to classify the input vector given. If some input vectors has the short distance then the input vector will be grouped into the same class. The LVQ method can be used to classify the data into some classes or categories. At this paper, the LVQ method will be applied to classify if someone is suffer potenciate of heart desease or not. The data that be trained are 268 data of heart desease patient from UCI (University of California at Irvine) with 10 variables that are factors influence that infected of heart desease. From some trials showed that the learning rate (α) = 0.25, decrease of learning rate (Decα) = 0.1, and the minimum learning rate (Minα) = 0.001 are values that give a good prediction with level of accuracy is about 66.79 %.   Keywords: Learning Vector Quantization, Classify, Heart Desease
PEMODELAN GENERAL REGRESSION NEURAL NETWORK UNTUK PREDIKSI TINGKAT PENCEMARAN UDARA KOTA SEMARANG Warsito, Budi; Rusgiyono, Agus; Amirillah, M. Afif
MEDIA STATISTIKA Vol 1, No 1 (2008): Media Statistika
Publisher : Jurusan Statistika FSM Undip

Show Abstract | Original Source | Check in Google Scholar | Full PDF (86.16 KB)

Abstract

This paper is discuss about General Regression Neural Network (GRNN) modelling to predict time series data, i.e. the air pollution rate in Semarang City comprises the floating dust, carbon monoxide (CO) and nitrogen monoxide (NO). The GRNN model have four processing layer that are input layer, pattern layer, summation layer and output layer. The input variable is determined by the ARIMA model. The result of GRNN modelling shows that the network have a good performance both at predict in sample and predict out of sample, that can be seen from the mean square error.   Keywords: GRNN, predict, air pollution  
PELATIHAN FEED FORWARD NEURAL NETWORK MENGGUNAKAN ALGORITMA GENETIKA DENGAN METODE SELEKSI TURNAMEN UNTUK DATA TIME SERIES Yuliandar, David; Warsito, Budi; Yasin, Hasbi
Jurnal Gaussian Vol 1, No 1 (2012): Wisuda Periode Oktober 2012
Publisher : Jurusan Statistika UNDIP

Show Abstract | Original Source | Check in Google Scholar

Abstract

ABSTRAK Pemodelan time series seringkali dikaitkan dengan proses peramalan suatu nilai karakteristik tertentu pada periode mendatang. Salah satu metode peramalan yang berkembang saat ini adalah menggunakan artificial neural network atau yang lebih dikenal dengan neural network.Penggunaan neural network dalam peramalan time series dapat menjadi solusi yang baik, namun yang menjadi masalah adalah arsitektur jaringan dan pemilihan metode pelatihan yang tepat. Salah satu pilihan yang mungkin adalah menggunakan algoritma genetika. Algoritma genetika adalah suatu algoritma pencarian stokastik berdasarkan cara kerja melalui mekanisme seleksi alam dan genetik yang bertujuan untuk mendapatkan solusi dari suatu masalah. Algoritma ini dapat digunakan sebagai metode pembelajaran dalam melatih model feed forward neural network. Penerapan algoritma genetika dan neural network untuk peramalan time series bertujuan untuk mendapatkan bobot-bobot yang optimum dengan meminimumkan error. Dari hasil pelatihan dan pengujian pada data kurs Dolar Australia terhadap Rupiah didapatkan nilai RMSE sebesar 117.3599 dan 82.4917. Model ini baik untuk digunakan karena memberikan hasil prediksi yang cukup akurat yang ditunjukkan oleh kedekatan target dengan output.
ANALISIS DATA RUNTUN WAKTU DENGAN METODE ADAPTIVE NEURO FUZZY INFERENCE SYSTEM (ANFIS) Saputra, Arsyil Hendra; Tarno, Tarno; Warsito, Budi
Jurnal Gaussian Vol 1, No 1 (2012): Wisuda Periode Oktober 2012
Publisher : Jurusan Statistika UNDIP

Show Abstract | Original Source | Check in Google Scholar

Abstract

One popular method of time series analysis is ARIMA. The ARIMA method requires some assumptions; residual of model must be white noise, normal distribution and constant variance. The ARIMA model tends to be better for time series data which is linear. Whereas for the nonlinear time series data have been widely studied by nonlinear methods, one of that is Adaptive Neuro Fuzzy Inference System or ANFIS. The ANFIS method is a method that combines techniques Neural Network and Fuzzy Logic. In this thesis discussed the ANFIS method specifically for the analysis of time series data that have characteristics such as stationary, stationary with outlier, non stationary and non stationary with outlier, and the data of Indonesian palm oil prices is used as a case study. The ANFIS results which were obtained are compared with the results of ARIMA method by the value of RMSE. Based on the analysis and discussion, it is obtained that the results of ANFIS method are better than the results of ARIMA method.
ANALISIS INTERVENSI DAN DETEKSI OUTLIER PADA DATA WISATAWAN DOMESTIK (Studi Kasus di Daerah Istimewa Yogyakarta) Budiarti, Lenny; Tarno, Tarno; Warsito, Budi
Jurnal Gaussian Vol 2, No 1 (2013): Wisuda Periode Januari 2013
Publisher : Jurusan Statistika UNDIP

Show Abstract | Original Source | Check in Google Scholar

Abstract

The tourist data is very interesting to be studied because the Indonesian tourism sector is an activator of the national economic which is potential to push higher economic growth in the future. Therefore, the forecast about tourist data is very needed for tourism business. The tourist data tend to fluctuate caused by many factors that affect the number of tourists extremely in an area, such as disasters, government regulation, social stability, violence and terrorism. That the extreme data can be assessed using intervention analysis and outlier detection. Intervention model is a time series model that can be used to forecast data consist of intervention of internal and external factors. In the intervention model, there are two kinds of intervention function, i.e., step and pulse functions. Step function is a form of intervention occurred in period of time while the pulse function is a form of intervention occurred only in a certain time. For the outlier detection, there are four types, such as additive outlier (AO), innovational outlier (IO), level shift (LS) and temporary change (TC). As an empirical studies was conducted by the domestic tourists data in Yogyakarta from January 2006 until December 2010 who staying on five-star hotels and motel throughout Yogyakarta. Based on the result of this research, known that the intervention occurred on January 2010 using the pulse function with MSE value 1172. Meanwhile based on the outliers detection, known any five outliers but only four outliers that significant included to the intervention model with MSE value 523,7167. So, the intervention model and outlier detection are chosen as a the best model based on the smallest MSE criterion. Keywords: Domestic tourists, intervention model, pulse function, outlier detection
METODE PERAMALAN DENGAN MENGGUNAKAN MODEL VOLATILITAS ASYMMETRIC POWER ARCH (APARCH) Elvitra, Cindy Wahyu; Warsito, Budi; Hoyyi, Abdul
Jurnal Gaussian Vol 2, No 4 (2013): Wisuda Periode Oktober 2013
Publisher : Jurusan Statistika UNDIP

Show Abstract | Original Source | Check in Google Scholar

Abstract

Exchange rate can be defined as a ratio the value of currency. The exchange rate shows a currency price, if it exchanged with another currency. Exchange rates of a currency fluctuate all the time. Rise and fall exchange rates of a currency in the money market shows the magnitude of volatility occurred in a country currency to other´s. To estimate the volatility behavior of the data gave rise to volatility clustering or heteroscedasticity problems, can’t be modeled using ARMA model and asymmetric effects that can‘t be modeled by ARCH or GARCH, can be modeled by Asymmetric Power ARCH (APARCH). In determining the estimated parameter values of APARCH model, used the maximum likelihood method, followed by using the iteration method is Berndt, Hall, Hall and Hausman (BHHH). The APARCH model used to the data return of exchange rate against dollar is APARCH(2,1) or in the form as follows :  = 0,00000268 + 0,830902 + 0,130516  + 0,074784  + 0,151157
PEMODELAN MARKOV SWITCHING AUTOREGRESSIVE Ariyani, Fiqria Devi; Warsito, Budi; Yasin, Hasbi
Jurnal Gaussian Vol 3, No 3 (2014): Wisuda Periode Agustus 2014
Publisher : Jurusan Statistika UNDIP

Show Abstract | Original Source | Check in Google Scholar

Abstract

Transition from depreciation to appreciation of exchange rate is one of regime switching that ignored by classic time series model, such as ARIMA, ARCH, or GARCH. Therefore, economic variables are modeled by Markov Switching Autoregressive (MSAR) which consider the regime switching. MLE is not applicable to parameters estimation because regime is an unobservable variable. So that filtering and smoothing process are applied to see the regime probabilities of observation. Using this model, transition probabilities and duration of the regime can be informed. In this case conducted exchange rate of Rupiah to US Dollar modeling with MSAR. The best model is MS(2)-AR(1) with transition probabilities from depreciation to appreciation is 0,052494 and appreciation to depreciation is 0,746716. Duration of the depreciation state is 19,04986 days and appreciation state is 1,339198 days.
PEMODELAN MARKOV SWITCHING VECTOR AUTOREGRESSIVE (MSVAR) Permatasari, Hayuk; Warsito, Budi; Sugito, Sugito
Jurnal Gaussian Vol 3, No 3 (2014): Wisuda Periode Agustus 2014
Publisher : Jurusan Statistika UNDIP

Show Abstract | Original Source | Check in Google Scholar

Abstract

Economic and financial variables are variables that are fluctuated because of regime switching as a result of political and economical conditions. Linear modeling can not capture the regime switching, so it is better to use Markov Switching Vector Autoregressive Models (MSVAR). MSVAR is a combination of vector autoregressive models and hidden markov models. Daily return of Rupiah buying rate against the USD and Euro are economic variables that are fluctuated and they can explain economic condition of a country. The best model of five order iteration is MS (2) - VAR (4) with the smallest AIC value, that is -1460.48.  Maximum Likelihood Estimation is a method to get parameters estimation. With 73 data, the return rates has transition probability 0.08 from crisis to normal state, while the transisition probablity of the opposite condition is 0.6. Expected value being at normal state is 13.10 days and being at crisis state is 1,68 days.
PEMODELAN TINGKAT INFLASI INDONESIA MENGGUNAKAN MARKOV SWITCHING AUTOREGRESSIVE CONDITIONAL HETEROSKEDASTICITY Wahyudi, Omy; Warsito, Budi; Prahutama, Alan
Jurnal Gaussian Vol 4, No 1 (2015): Wisuda Periode Januari 2015
Publisher : Jurusan Statistika UNDIP

Show Abstract | Original Source | Check in Google Scholar

Abstract

The financial sector often under conditions of fluctuating due to changes in monetary policy, the political instability even just a rumor. The linear model cannot capture changes in these conditions, so the model used is Markov Switching Autoregressive Conditional Heteroskedasticity (SWARCH). This model produces value of transition probability and the duration of each state. Filtering and smoothing process performed to determine probability of the observation data in each state. Modeling about the inflation data in Indonesia was done. The model used is SWARCH (2.1) with 240 data. The probability of inflation rate switch from non crisis state to crisis state is 0.016621, while the probability of inflation rate switch from crisis state to non crisis state is 0.195719. Expectation value of the length time in non crisis state is 60.16 days and the crisis state is 5.11 days.Keywords :  filtering, smoothing, transition probability, SWARCH
PREDIKSI CURAH HUJAN KOTA SEMARANG DENGAN FEEDFORWARD NEURAL NETWORK MENGGUNAKAN ALGORITMA QUASI NEWTON BFGS DAN LEVENBERG-MARQUARDT Warsito, Budi; Sumiyati, Sri
JURNAL PRESIPITASI Vol 3, No 2 (2007): Vol 3, No 2 (2007)
Publisher : Department of Environmental Engineering

Show Abstract | Original Source | Check in Google Scholar | Full PDF (120.768 KB)

Abstract

This paper study the rainfall prediction at Semarang City as time series data with Feed Forward Neural Network  (FFNN)  model.  The  learning  algorithm  that  be  used  are  the  Quasi  Newton BFGS and Levenberg-Marquardt algorithm. The input unit is determined based on the best of ARIMA model. The computation is done with use  Matlab 7.1 program with 1000 epoch, five unit of hidden layer, 100 replication  and use  input at lag  variabel  1,  12  and 13, respectively. The result shows that the prediction is good in relatively, where Quasi Newton BFGS algorithm result the Mean Square Error (MSE) that more accurate.