ANALYZING THE IMPACT OF DATA AUGMENTATION ON HANDWRITING BASELINE PATTERN DETECTION PERFORMANCE USING THE YOLO FRAMEWORK

Abstract

Handwriting baseline pattern detection is a critical step in handwriting analysis, as it reflects the directional tendencies of writing strokes. However, class imbalance in the dataset often leads to degraded model performance on minority classes. This study aims to analyze the impact of data augmentation on detection performance using YOLOv11. The dataset consists of 150 handwriting images containing 1,432 baseline instances, classified into Ascending Baseline (AB) with 530 instances, Descending Baseline (DB) with 611 instances, and Wavy Baseline (WB) with 291 instances. Experimental results show that augmentation substantially improved Precision from 0.383 to 0.517, F1-Score from 0.505 to 0.596, mAP@50 from 0.503 to 0.614, and mAP@50–95 from 0.355 to 0.470. Additionally, the detection rate for the WB class, derived from the normalized confusion matrix, increased from 4% to 37%. These findings confirm that data augmentation has a significant positive impact on handwriting baseline pattern detection using YOLOv11, and contribute to the advancement of AI-driven detection methodologies.