Improving Accuracy in Fingerprint Attribute Classification Using Convolutional Neural Network Compared to Resnet50 Algorithm
Sohini Kamal.G, Ramesh Shanmugam, Najm Us Sama · 2024
The primary objective of this study is to enhance the accuracy of fingerprint attribute classification. To achieve this, the research utilized a dataset from Kaggle. The study comprised two groups: Group I and Group II, each consisting of 20 samples. Group I employed a Convolutional Neural Network (CNN), while Group II utilized the ResNet50 architecture, resulting in a total sample size of$\mathbf{4 0}$. Sample size calculations for statistical analysis were conducted, and performance comparisons were implemented using Python. The statistical analysis was performed on clincalc.com, with a statistical power (G-power) of$\mathbf{8 5 \%}$, an alpha level$\alpha$of$\mathbf{0. 0 5}$, and a beta$\beta$of$\mathbf{0. 2}$. The main focus of the analysis was to compare the performance of the CNN and ResNet50 using accuracy as the primary evaluation metric. The results revealed that the Convolutional Neural Network achieved an accuracy of$\mathbf{9 3. 8 5 9 \%}$, significantly outperforming ResNet50, which recorded an accuracy of 81.052 %. The statistical significance was confirmed with a two-tailed$p$-value of less than 0.001, indicating strong evidence against the null hypothesis. In conclusion, the Convolutional Neural Network demonstrates superior accuracy compared to ResNet50 in the context of fingerprint attribute classification.