<![CDATA[This study proposes an OCR–LSTM-based framework for hazardous cosmetic ingredient detection and skin-type classification using ingredient analysis. The increasing use of skincare products has raised significant safety concerns, particularly regarding the presence of harmful substances in products distributed through online platforms. To address this issue, the proposed framework integrates Optical Character Recognition (OCR) for extracting ingredient text from cosmetic labels with a Long Short-Term Memory (LSTM) model for classification and analysis.Two primary datasets were utilized: a hazardous ingredient list obtained from the Indonesian Food and Drug Authority (BPOM) and a skincare ingredient dataset categorized based on compatibility with different skin types, including oily, dry, normal, and combination. The extracted text was processed through preprocessing and normalization stages before being used as input for the LSTM model to classify ingredient safety and determine skin-type suitability. Experimental results show that the proposed framework achieved a training accuracy of 97.59% and a validation accuracy of 96.68%, with strong classification performance, particularly for Combination, Normal, and Dry skin categories. These results demonstrate that the integration of OCR and deep learning provides an effective approach for automated ingredient analysis, enabling accurate detection of hazardous substances and supporting safer skincare product selection.]]>
Copyrights © 2026
