Fraud Prediction Model on Premium Cosmetics Transactions Using Deep Learning: A Long Short-Term Memory (LSTM) Approach
Abstrak
The rapid growth of the premium cosmetics industry has significantly increased online and offline transactions, but also heightened the risk of fraud. Traditional detection approaches often fail to capture dynamic patterns. This study proposes a fraud prediction model using Long Short-Term Memory (LSTM), a deep learning architecture suitable for sequential transaction data. Unlike previous studies that mainly focus on banking and general e-commerce fraud, this research specifically addresses premium cosmetics transactions, a domain with limited exploration. The dataset consists of 2,133 transactions with 16 features covering demographics, transaction details, and technical attributes. After preprocessing (cleaning, normalization, categorical encoding, and sequential arrangement), the LSTM model was trained and validated (70-15-15 split), achieving 94.2% accuracy, 91.5% precision, 89.7% recall, 90.6% F1-score, and 0.95 AUC. These results highlight the novelty and effectiveness of LSTM in detecting fraudulent patterns in the premium cosmetics sector, offering practical implications for enhancing security and trust in high-value transactions.
Identifier
- DOI: https://doi.org/10.20895/centive.v2025i1.519
- OpenAlex: https://openalex.org/W7127098426