Ummah, Nurul Wakhidatul (2016) Deteksi Anomali Pada Proses Bisnis Menggunakan Metode Multi Criteria Decision Making (MCDM) Dan Hybrid Neural Network. Undergraduate thesis, Institut Teknologi Sepuluh Nopember Surabaya.
Preview |
Text
5111100003-Undergraduate_Thesis.pdf - Published Version Download (3MB) | Preview |
Abstract
Fraud merupakan suatu kecurangan yang menyebabkan
kerugian besar bagi perusahaan. Fraud dapat dideteksi dengan melihat
anomali yang ada pada proses bisnis perusahaan. Eksekusi waktu yang
tidak sesuai, aktivitas resource yang tidak sesuai dengan wewenangnya,
aktivitas yang sengaja dilewati, dan pengambilan keputusan yang salah
merupakan beberapa contoh anomali yang sering terjadi. Studi kasus
tersebut dapat menggunakan ontologi untuk mendapatkan nilai atribut,
kemudian dilakukan perhitugan rating menggunakan metode MCDM dan
label diperoleh dari verifikasi pakar dan nilai fuzzy triangular. Label akan
menjadi masukan untuk metode hibrid neural network yang berfungsi
mendeteksi anomali. Metode ini merupakan pengembangan dari neural
network dengan penggunaan fungsi Gaussian di hidden layer untuk
mengatasi masalah aproksimasi dari value of attributes dan penambahan
ridge sebagai solusi masalah multikolinearitas atribut anomali yang
belum bisa diatasi oleh metode neural network sebelumnya. Uji coba
metode digunakan 1200 dataset dengan cross validation dan kombinasi
lima parameter. Hasil klasifikasi yang paling optimal menggunakan
minimum seed 2, maximal iteration tak hingga, minimal standar deviasi
0,02, jumlah clustering 3, dan nilai ridge dari RBF Network 1,0 e-7.
Performa yang dihasilkan sangat baik yakni akurasi 99%, presisi 100%,
sensitivitas 71%, dan spesivisitas 100%. Oleh karena itu metode hybrid
neural network dapat digunakan untuk mendeteksi anomali
============================================================================================
Fraud is a fraudulent that results a huge financial loss to
companies. Fraud is able to be detected by the anomaly shown in the
business process. The inappropriate time execution, unauthorized
resource activities, intentionally by pass activity, and wrong decision
making are the examples of frequent anomaly. From the aforementioned
cases, the ontology method can be used to gain the attributes value. Then,
the rating can be calculated the MCDM method as well as the label is
gained from the expert verification and the value of triangular fuzzy. The
label will be an input for hybrid neural network method which has a
function to detect the anomaly. This method was the development of
neural network with the addition of Gaussian function in the hidden layer
to solve the approximation problem from values of attributes and the
ridge as a solution of the anomaly attributes multicolinearity problem that
unable to be overtaken by the previous neural network method. This
method was tested using 1200 dataset with cross validation and the
combination of five parameters. The most optimum clasification was
using the minimum seed of 2, infinity maximal iteration, minimal
deviation standard of 0.02, the total clustering of 3, as well as the value
of ridge from RBF Network of 1.0e-7. This performance had a good
accurate number which is 99%, presision 100%, sensitivity 71%,
specificity 100%. Therefore this hybrid neural network method can be
used to detect the anomaly
| Item Type: | Thesis (Undergraduate) |
|---|---|
| Additional Information: | RSIf 658.403 801 1 Umm d |
| Uncontrolled Keywords: | Anomali, Ontologi, MCDM, Hybrid Neural Network |
| Subjects: | T Technology > T Technology (General) > T58.6 Management information systems |
| Divisions: | Faculty of Information Technology > Informatics Engineering > 55201-(S1) Undergraduate Thesis |
| Depositing User: | EKO BUDI RAHARJO |
| Date Deposited: | 10 Jul 2020 02:36 |
| Last Modified: | 10 Jul 2020 02:36 |
| URI: | http://repository.its.ac.id/id/eprint/76349 |
Actions (login required)
 |
View Item |