Deteksi outlier merupakan salah satu penelitian yang penting dengan tujuan mencari data yang memiliki karakteristik yang berbeda dengan kebanyakan data lainya. Data outlier yang ada di dataset akan mempengaruhi terhadap hasil dari proses selanjutnya seperti clustering dan klasifikasi dan terkadang outlier juga menyimpan informasi yang penting. Penelitian ini dimulai dengan pemotongan daerah penting pada citra yang dilakukan secara manual. Setelah mendapatkan citra dari proses sebelumnya, kemudian dilakukan ektraksi fitur warna dengan perhitungan statistik orde pertama yaitu mean, standard deviation, kurtosis dan skewness dan ektraksi fitur tekstur menggunakan algoritma Grey Level Co-Occurrence Matrix (GLCM) dengan jarak 1 dan sudut 0o dengan fitur yang digunakan yaitu Angular Second Moment, Contras, Inverse Different Moment, Cluster Shade, dan Cluster Prominence. Setelah didapatkan fitur-fitur setiap citra kemudian dilakukan clustering dengan menggunakan Algoritma Clustering Large Applications Based on Randomized Search (CLARANS). Setelah proses clustering kemudian dilakukan deteksi outlier dengan menggunakan Algoritma unweighted-CBLOF. Dataset yang digunakan pada penelitian ini adalah dataset Supermarket Produce dengan jumlah 2633 citra sebagai data normal dan data outlier sebanyak 50 diambil dari dataset Fruit Images. Hasil dari penelitian ini didapatkan fitur yang terbaik untuk deteksi outlier yaitu menggunakan fitur tekstur Grey Level Co-Occurrence Matrix (GLCM) dengan nilai akurasi sebesar 85,612 % dan penentuan jumlah kelas pada Clustering Large Applications Based on Randomized Search (CLARANS) yang semakin meningkat belum tentu meningkatkan nilai akurasi, sehingga perlu dilakukan uji coba dengan beberapa jumlah kelas untuk mencari nilai akurasi yang paling tinggi.

Outlier detection is one of many important researches which aimed to search for data whose its characteristics differs with other commonly data. Outlier data which was in dataset could influence the result of its next process, for example, clustering and classification and somehow we also had to know that outlier also saved some important information. This research started at truncation of momentous area into the images, and it was done manually. After we got the images from this process, the next one was color feature extraction method, where this process used first-order statistical in image processing, they were: mean, standard deviation, kurtosis and skewness. Meanwhile, it was also necessary to did color feature extraction method which used Grey Level Co-Occurrence Matrix (GLCM) algorithm within distance 1 and angle 0o. This process used some features, which were: Angular Second Moment, Contras, Inverse Different Moment, Cluster Shade and Cluster Prominence. After we got these every single features from the images, the next process was clustering which used Clustering Large Application Based on Randomized Search (CLARANS) algorithm. After this clustering process, the next one was outlier detection within Algorithm unweighted – CBLOF. Dataset which in used was Supermarket Produce dataset, within amount of 2633 images as a normal data. And 50 outlier data was got from Fruit images dataset. Result of this research was all of these processes could get the best features for outlier detection, where it used Grey Level Co-Occurrence Matrix (GLCM) texture features. And within its point of accuration was 85.612%. Meanwhile, about the total amount of class determination on Clustering Large Application Based on Randomized Search (CLARANS) which was higher and higher, it could not definitely also being a determinant to make a higher accuration score. So that, it was necessary to have a test and try within some classes, where this process aimed to search for the highest accuration score.

Baca Jurnal Baca Jurnal Baca Jurnal Baca Jurnal Baca Jurnal Baca Jurnal Baca Jurnal Baca Jurnal Baca Jurnal