<![CDATA[Abstract: Arabica coffee is a high-value export commodity for the Indonesian economy. To maintain global competitiveness, coffee beans must meet export feasibility quality standards based on the cupping score from the Coffee Quality Institute (CQI). This study aims to compare the performance of the K-Nearest Neighbor (KNN) and Gaussian Naive Bayes algorithms in classifying the export feasibility of Arabica coffee beans. In sensory quality testing, high-dimensional attributes (10 parameters) can cause accuracy instability and increase computational load. Therefore, Correlation-based Feature Selection (CFS) was applied to reduce redundant features. The CFS process filtered the initial 10 features into 4 selected features (Flavor, Acidity, Cupper Points, and Aroma). The Dataset was obtained from the Kaggle Coffee Quality Database and had undergone Excel format adjustments before being loaded into the system. Modeling was conducted using 1,303 data records with a 30% split for testing data. The evaluation results showed that the Naive Bayes algorithm provided the best performance, achieving an accuracy rate of 97.95%. The use of CFS proved successful in reducing feature dimensions by 60% without significantly decreasing classification accuracy. Keywords: Arabica Coffee; Export Classification; K-Nearest Neighbor; Naive Bayes; Correlation-based Feature Selection. Abstrak: Kopi Arabika merupakan komoditas ekspor bernilai tinggi bagi perekonomian Indonesia. Untuk menjaga daya saing di tingkat global, biji kopi harus memenuhi standar kualitas kelayakan ekspor berdasarkan cupping score dari Coffee Quality Institute (CQI). Penelitian ini bertujuan membandingkan kinerja algoritma K-Nearest Neighbor (KNN) dan Gaussian Naive Bayes dalam klasifikasi kelayakan ekspor biji kopi Arabika. Dalam pengujian sensoris mutu, dimensionalitas atribut yang tinggi (10 parameter) dapat menyebabkan ketidakstabilan akurasi dan meningkatkan beban komputasi. Oleh karena itu, Correlation-based Feature Selection (CFS) diterapkan untuk mereduksi fitur redundan. Proses CFS menyaring 10 fitur awal menjadi 4 fitur terpilih (Flavor, Acidity, Cupper Points, Aroma). Dataset diambil dari Kaggle Coffee Quality Database dan telah melalui tahapan penyesuaian format Excel sebelum dimuat ke sistem. Pemodelan dilakukan menggunakan 1303 rekam data dengan pembagian data uji sebesar 30%. Hasil evaluasi menunjukkan bahwa algoritma Naive Bayes memberikan performa terbaik dengan tingkat akurasi mencapai 97,95%. Penggunaan CFS terbukti berhasil memangkas dimensi fitur sebesar 60% tanpa menurunkan akurasi klasifikasi secara signifikan. Kata Kunci: Kopi Arabika; Klasifikasi Ekspor; K-Nearest Neighbor; Naive Bayes; Correlation-based Feature Selection.]]>
Copyrights © 2026
