@mastersthesis{digilib71855,
           month = {March},
           title = {ANALISIS PERBANDINGAN KLASIFIKASI CITRA VARIAN GITAR AKUSTIK MENGGUNAKAN ALGORITMA CNN DAN KNN},
          school = {UIN SUNAN KALIJAGA YOGYAKARTA},
          author = {NIM.: 23206051004 Dafid Marwan Anggara},
            year = {2025},
            note = {Dr. Ir. Aulia Faqih Rifa'i, M.Kom},
        keywords = {CNN, KNN, Dataset, Citra},
             url = {https://digilib.uin-suka.ac.id/id/eprint/71855/},
        abstract = {The rapid advancement of digital image processing technologies has significantly contributed to the emergence of sophisticated object classification techniques. This study aims to compare the performance of two machine learning algorithms{--}Convolutional Neural Network (CNN) and K-Nearest Neighbors (KNN){--}in classifying acoustic guitar variants based on image data. CNN is known for its ability to recognize complex visual patterns, while KNN is more efficient for small-scale datasets with simpler structures. The research employed several preprocessing steps including image resizing, normalization, and augmentation to prepare a dataset of 1,500 images categorized into Auditorium, Classical, and Dreadnought types. CNN automatically extracts features using convolutional layers, whereas KNN relies on manual feature extraction using Histogram of Oriented Gradients (HOG). Performance evaluation was conducted using accuracy, precision, recall, F1-score, and processing time. The findings revealed that CNN outperformed KNN with a classification accuracy of 74\% compared to 71\%, and also achieved higher scores in other evaluation metrics. This superiority is attributed to CNN?s capability to learn intricate visual features, although it requires longer training time and more computational resources. Conversely, KNN is easier to implement and faster, but less effective for larger datasets. The study concludes that CNN is better suited for complex, large-scale image classification tasks, while KNN remains viable for simpler applications. Recommendations for future work include expanding the dataset, testing alternative algorithms such as SVM or Random Forest, and combining visual and auditory guitar features for a more comprehensive approach. The proposed models have potential applications in e-commerce platforms for automated product identification and verification.}
}