Search

This study aimed to know the effectiveness of oblique light for document examination in detecting indented writing on different types of papers and to study the properties of writing through the detected indented writing. Quantitative research was used, to determine that the oblique light source used to establish its effectiveness came from the left and right sides, ten millimeters above the sheet of paper, and the oblique light source at an angle of 10° to the level line. The study used different papers. - 70 and 80 gsm copy paper and 70 gsm report paper. It also studied the influence of pen types used in writing using gel ink pens and ballpoint pens, which, according to studies, showed that oblique light inside the proof of document examination a maximum of 3 sheets of indented writing on parchment sheets. Ballpoint pens writting on 70 gsm, 80 gsm, and report papers averaged 1.5, 1.2, and 1.2, respectively. Gel pens writting in 70 gsm, 80 gsm paper and report paper in the ink pen averaged 1.0, 1.0, and 0.8, respectively. Although the average number of sheets of paper detected for indented writing was different, according to the analysis of two variances factored at a statistical significance of p<0.05, different types of paper and pens did not affect the effectiveness of detecting pressure marks with a poof of document examination. The research will be useful in examining documentary evidence in forensic science. [ABSTRACT FROM AUTHOR]

Generative Adversarial Networks (GANs) are a class of deep neural networks that can be used to generate data examples in imbalanced data situations. GANs consist of two simultaneously trained modes: generative and discriminative modeling. The generative model generates new data as random noise from the training dataset, and the discriminator model distinguishes examples from generated new data and real data. We studied the overlapping data transfer during a generating model in distributed real-time data streaming. This paper proposes a new extension method on GANs called GANs2T based on tabular time series to improve the model performance and execution time. We use this technique to capture the covariance structure of the minority class and to generate synthetic samples along the probability contours for learning algorithms on streaming data. The experimental testing is performed on binary-class and multi-class imbalanced learning methods from several benchmark datasets. The results validate GANs2T combine with the XGBoost (GANs2T+XGBoost) algorithm for the overall accuracy = 84.93%, precision = 90.48%, recall = 88.13%, F1-score = 89.93 and average training time for training model = 60.20 seconds. [ABSTRACT FROM AUTHOR]