<![CDATA[This study presents a multi-device image collection and a reproducible VSCode/Python pipeline for analyzing image classification and the effects of data augmentation under real hardware variation. Images were captured at the Galeri Inovasi Institut Teknologi Sepuluh Nopember (GRITS) using four devices Infinix Note 4, LG G6, Samsung S23+, and Xiaomi Pad 6s Pro with 31 images per device. We applied manual and Python-based augmentations (rotation, flips, brightness, sharpening, contrast) and organized outputs by device and augmentation type for controlled comparisons. Using stratified 80:20 splits, we evaluated Logistic Regression (LR), SVM (RBF), and KNN. Results: LR reached accuracy 0.90 (macro-F1 0.88; weighted-F1 0.90), SVM 0.89 (macro-F1 0.88; weighted-F1 0.89), and KNN 0.67 (macro-F1 0.65; weighted-F1 0.68). Augmentation enhanced robustness and cross-device generalization, though Xiaomi Pad 6s Pro remained the most challenging class, indicating a persistent device-specific shift. The dataset and scripts provide a transparent, baseline-ready testbed for research on image classification, cross-device variability, and the impact of augmentation.]]>
Copyrights © 2025
