<![CDATA[The purpose of this study is to create and contrast image classifiers which are able to identify clean clothing from dirty using machine learning and deep learning techniques. Our utilized dataset contains 200 images of an outfit which are obtained from different official fashion brand sites and well-known e-commerce platforms in Indonesia. The information is used from secondary data which are digital images of the two style categories. Image preprocessing (resizes, normalizations and data augmentations), feature extraction from VGG-16, VGG-19 and inception v3 is done. The extracted features are then fed into the classifiers namely Logistic Regression, Neural Network and Support Vector Machine (SVM). The evaluation of the model is performed by different metrics (e.g., AUC, accuracy, F1-score, precision, recall and MCC) and visual examination using MDS plot and Silhouette Plot. The results demonstrate that the integrated model involving VGG-16 and Logistic Regression performs best obtaining highest AUC when compared with other model combinations. The MDS and Silhouette Plot visualizations also supported that VGG-16 has the most superior feature separation between clean outfits and non-clean outfits. In a word, our study unveils that fashion style recognition accuracy can be improved significantly through CNN-based feature extraction and traditional classification model. We hope that our work will encourage the comparison of CNN feature extraction and classification algorithms, and also can lay the foundation for further research in image-based outfit guidance systems serving a range of fashion industry and service sectors where professional appearance is a criterion.]]>
