@phdthesis{digilib78704,
           month = {August},
           title = {PEMILIHAN MODEL REGRESI LOGISTIK LASSO MENGGUNAKAN AIC: STUDI KASUS PREDIKSI DEFAULT KREDIT},
          school = {UIN SUNAN KALIJAGA YOGYAKARTA},
          author = {NIM.: 22106010087 Lucky Ramadhani},
            year = {2026},
            note = {Sri Utami Zuliana, S.Si., M.Sc., Ph.D.},
        keywords = {Credit Scoring, Regresi Logistik, LASSO, AIC, Risiko Gagal Bayar},
             url = {https://digilib.uin-suka.ac.id/id/eprint/78704/},
        abstract = {Credit default risk poses a major challenge in the banking industry, which can
be mitigated through predictive credit scoring models. This study predicts credit card
default risk using Least Absolute Shrinkage and Selection Operator (LASSO) logistic
regression optimized by the Akaike Information Criterion (AIC). The LASSO method
eliminates redundant and highly correlated transaction variables (multicollinearity),
while AIC selects the simplest and most accurate model architecture. Utilizing 30,000
customer records from the UCI Machine Learning Repository partitioned into 80\%
training and 20\% testing sets, the algorithm successfully removed two redundant
billing variables (BILL AMT4 and BILL AMT5) at the optimal penalty level of {\ensuremath{\lambda}} =
0.0002013. The final model achieved an overall accuracy of 81.16\%, a specificity of
97.20\% in identifying non-default customers, and an AUC discrimination power of
0.7202. Furthermore, risk factor analysis revealed that the most recent repayment delay
(PAY 1) is the most critical trigger, multiplying default risk by 1.766 times per delay
level. This study concludes that integrating LASSO with AIC yields a concise, stable,
and practical credit scoring system to assist banks in detecting high-risk borrowers
early.}
}