<![CDATA[Ketidak seimbangan data biologis lapangan dan kompleksitas morfologi alami akibat sifat hermafrodit protandri merupakan tantangan utama dalam pemantauan reproduksi komoditas perikanan laut lepas. Penelitian ini bertujuan untuk mengoptimasi akurasi prediksi Tingkat Kematangan Gonad (TKG) udang utara (Pandalus borealis) melalui rekonstruksi data berbasis pendekatan Knowledge Discovery in Databases (KDD) dan kecerdasan komputasional. Metode penelitian diawali dengan merestrukturisasi lima kategori kematangan asli NOAA menjadi tiga kelas target yang tegas, yaitu Belum Matang, Peralihan, dan Matang. Ruang data tabular dinormalisasi menggunakan skala RobustScaler untuk meredam pengaruh pencilan (outliers), sementara masalah ketimpangan kelas yang ekstrem ditangani secara hibrida menggunakan kombinasi algoritma SMOTE dan Tomek Links pada data latih. Pemodelan klasifikasi dibangun menggunakan arsitektur Tree-Based Ensemble Learning tingkat lanjut yang menyinergikan tiga algoritma dasar (XGBoost, LightGBM, dan Random Forest) melalui metode Stacking dengan meta-learner Logistic Regression berbasis fungsi Softmax. Hasil pengujian menggunakan 20% data uji independen menunjukkan performa model LightGBM memberikan performa terbaik dengan capaian metrik akurasi sebesar 90,03% dan nilai F1-Score Macro sebesar 87,08%. Melalui analisis Permutation Feature Importance, model memberikan transparansi fungsional dengan menetapkan variabel umur (AGE dan LOG_AGE) serta jenis kelamin (SEX) sebagai dua fitur biometrik paling berpengaruh dari total 31 fitur hasil rekayasa data. Kesimpulan dari penelitian ini menegaskan bahwa implementasi model Ensemble berbasis penyeimbangan data hibrida memiliki urgensi penting untuk diaplikasikan dalam digitalisasi manajemen stok dan penentuan regulasi musim tangkap krustasea secara berkelanjutan. The imbalance of field biological data and the complexity of natural morphology due to the protandrous hermaphrodite nature are major challenges in monitoring the reproduction of open sea fisheries commodities. This study aims to optimize the accuracy of the Gonad Maturity Level (GMT) prediction of northern shrimp (Pandalus borealis) through data reconstruction based on the Knowledge Discovery in Databases (KDD) approach and computational intelligence. The research method begins by restructuring the five original NOAA maturity categories into three clear target classes, namely Immature, Transitional, and Mature. The tabular data space is normalized using the RobustScaler to reduce the influence of outliers, while the problem of extreme class imbalance is handled in a hybrid manner using a combination of the SMOTE and Tomek Links algorithms on the training data. Classification modeling is built using an advanced Tree-Based Ensemble Learning architecture that synergizes three basic algorithms (XGBoost, LightGBM, and Random Forest) through the Stacking method with the Logistic Regression meta-learner based on the Softmax function. The test results using 20% of the independent test data showed a LightGBM model achieved the best performance, with an accuracy of 90.03% and a Macro F1-Score of 87.08%. Through Permutation Feature Importance analysis, the model provides functional transparency by determining age (AGE and LOG_AGE) and sex (SEX) as the two most influential biometric features from a total of 31 data engineering features. The conclusion of this study confirms that the implementation of the Ensemble model based on hybrid data balancing has a significant urgency to be applied in the digitalization of stock management and the determination of sustainable crustacean fishing season regulations. ]]>
Copyrights © 2026
