Enhancing Accuracy in Image Recognition for Deep Faked Images using Convolutional Networks Compared with Decision Tree Algorithm

This study's main goal is to improve deep fake image identification accuracy by using convolutional neural networks (CNN) and contrasting the outcomes with decision tree algorithms. Examining methods like feature extraction, transfer learning, and model evaluation can help deep fake detection systems become more reliable and effective. A dataset from Kaggle. For training and assessment, a combination of real and deep faked images is used. The dataset contains samples produced using different circumstances and methods for deep fake production. In this work, two different groups were implemented, each using a different algorithmic strategy for image identification. Group 1 utilized CNNs, or convolutional neural networks. Group 2 used the Decision Tree Algorithm, a hierarchical tree-like topology. With 20 samples in each group, there were 40 samples in total. In order to provide sufficient statistical power for further analyses, the sample size was determined. Using Clincalc.com, an a priori power analysis was performed with the statistical power (G-power) set at 80%. Behavioural science conventions were followed while choosing the significance level (alpha, a) and type II error rate (beta, β). The values selected were 0.2 and 0.05, respectively. A 95% confidence interval was chosen to assess the precision of the estimated effectiveness of the two algorithmic approaches. The accuracy of the Decision Tree was 79.2%, while the CNNs outperformed it with a 97.36% accuracy. A two-tailed significance test result showing a statistically significant p-value of 0.001 (p < 0.05) for accuracy further supported the CNN model's superiority over the Decision Tree model. This suggests that there is statistical significance to the observed variations in accuracy between the two models. Convolutional networks performed better than other models, which was explained by their skill in extracting features from pictures. To distinguish between real and deep faked images, the model must be able to recognize complex patterns and spatial correlations, which is made possible by the hierarchical convolutional layers. A priori power analysis was used to establish the sample size that was selected, which gave the study enough statistical power. The results' validity was reinforced by the 95% confidence interval, which demonstrated a high degree of accuracy in predicting the algorithmic approaches' efficacy. As seen by the significant gains in accuracy and precision measures, the study confirms that the novel CNN routinely outperforms the conventional feedforward decision tree algorithm in the field of deep fake image identification.