Electronic Theses and Dissertation

Penyalahgunaan kartu kredit merupakan ancaman serius dalam sektor perbankan seiring dengan meningkatnya volume transaksi digital. penelitian ini bertujuan untuk: (1) menerapkan algoritma light gradient boosting machine (lightgbm) guna mendeteksi anomali pada transaksi kartu kredit, dan (2) menganalisis hasil kinerja lightgbm berdasarkan metrik accuracy, precision, recall, f1-score, dan roc auc. data yang digunakan dalam penelitian ini bersumber dari dataset simulasi transaksi kartu kredit yang mencakup berbagai variabel aktivitas pengguna. hasil pengujian menunjukkan bahwa penerapan lightgbm sangat efektif dalam mengidentifikasi transaksi anomali dengan capaian akurasi sebesar 98,63% dan specificity 98,69%. analisis kinerja berdasarkan metrik evaluasi menghasilkan nilai recall sebesar 95,79%, precision 30,16%, dan f1-score 45,88%. meskipun nilai presisi terdampak oleh karakteristik data yang tidak seimbang (imbalanced data), model menunjukkan kemampuan pemisahan kelas yang sangat istimewa dengan skor roc auc mencapai 0,9973. kinerja tersebut diperoleh melalui konfigurasi hyperparameter dengan nilai num leaves = 31, learning rate = 0,05, dan n estimators = 500. selain itu, ditemukan bahwa variabel nominal transaksi (amt) merupakan faktor paling signifikan dalam pembentukan model. secara keseluruhan, penelitian ini menegaskan bahwa algoritma lightgbm dapat diandalkan sebagai instrumen deteksi dini untuk meminimalkan risiko kerugian akibat penipuan kartu kredit.

Credit card fraud represents a serious threat in the banking sector as the volume of digital transactions continues to rise. This study aims to: (1) apply the Light Gradient Boosting Machine (LightGBM) algorithm to detect anomalies in credit card transactions, and (2) analyze the performance of LightGBM based on the evaluation metrics of Accuracy, Precision, Recall, F1-Score, and ROC AUC. The data used in this study were sourced from a simulated credit card transaction dataset encompassing various user activity variables. The experimental results demonstrate that the application of LightGBM is highly effective in identifying anomalous transactions, achieving an accuracy of 98.63% and a specificity of 98.69%. Performance analysis based on evaluation metrics yielded a recall of 95.79%, precision of 30.16%, and F1-score of 45.88%. Although the precision value was affected by the imbalanced nature of the dataset, the model demonstrated an outstanding class separation capability with a ROC AUC score of 0.9973. These results were obtained through hyperparameter configuration with values of num_leaves = 31, learning_rate = 0.05, and n_estimators = 500. Furthermore, the transaction amount variable (amt) was identified as the most significant factor in the model construction. Overall, this study confirms that the LightGBM algorithm can be reliably employed as an early detection instrument to minimize the risk of financial losses due to credit card fraud