Image spam involves the practice of concealing text within an image. Various machine-learning techniques are used to categories image spam, utilizing a wide range of features extracted from the images. Convolutional neural networks (CNNs) are commonly used for image classification and feature extraction tasks because of their outstanding performance. In this study, our focus is to analyses image spam using a CNN model that incorporates deep learning techniques. This model has been meticulously fine-tuned and optimized to deliver exceptional performance in both feature extraction and classification tasks. In addition, we performed comparative evaluations of our model on different image spam datasets that were specifically created to make the classification task more challenging. The results we obtained show a significant improvement in classification accuracy compared to other methods used on the same datasets.
Read MoreDoi: https://doi.org/10.54216/JCIM.150106
Vol. 15 Issue. 1 PP. 62-76, (2025)
Steganography involves concealing hidden messages inside various types of media, whereas steganalysis is the process of identifying the presence of steganography. Convolutional neural networks (CNN), a type of neural network that outperformed previously proposed machine learning-based methods when introduced, are among the models used for deep learning. While CNN-based methods may yield satisfactory results, they face challenges in terms of classification accuracy and network training stability. The present research introduces a CNN structure to increase hidden data detection and spatial domain image training reliability. The suggested method includes pre-processing, feature extraction, and classification. Evaluation of performance is conducted on datasets Break Our Steganographic System Base (BOSSbase-.01) and Break Our Watermarking System (BOWS2) with three adaptive steganography algorithms. Wavelet Obtained Weights (WOW), Spatial Universal Wavelet Relative Distortion (S-UNIWARD), and Highly Undetectable steGO (HUGO) operating at low payload capacities of 0.2 and 0.4 bits per pixel (bpp). The experimental results surpass the accuracy and network stability of prior publications. Training accuracy ranges from 91% to 94%, and testing accuracy ranges from 74.8% to 86.65%.
Read MoreDoi: https://doi.org/10.54216/JCIM.150101
Vol. 15 Issue. 1 PP. 01-10, (2025)
Fake review detection, often known as spam review detection, is a crucial aspect of natural language processing. It involves extracting valuable information from text documents obtained from various sources. Various methodologies, such as simple rule-based approaches, lexicon-based methods, and advanced machine learning algorithms, have been extensively employed with diverse classifiers to provide accurate detection of fake reviews. Nevertheless, review classification based on lexicons continues to face challenges in achieving high accuracies, mostly because of the need for domain-specific comprehensive dictionaries. Furthermore, machine learning-driven review detection also addresses the limitations in accuracy caused by the uncertainty of features in social data. In order To address the problem of accuracy, one effective approach is to carefully choose the most optimal set of features and minimize the number of features used. The Objective of the research paper is to select a small subset of features out of the thousands of features for accurate classification of spam review detection. Therefore, a good feature selection method is needed in order to speed up the processing rate and predictive accuracy. This paper, Harris Hawks Optimization (HHO), is utilized for feature selection in sentiment analysis tasks. The performance of the selected feature subsets was evaluated using SVM, X-GBoost, ETC classifiers. Experimental results on tweet reviews for the airline dataset demonstrated superior sentiment classification capabilities, achieving an accuracy of 0.9435% with SVM and 0.9607%, 0.9635% for X-Boost, ETC, respectively.
Read MoreDoi: https://doi.org/10.54216/JCIM.150102
Vol. 15 Issue. 1 PP. 11-21, (2025)
The problem of data security in EHR is deeply concerning, as well as the methods used in session, feature, service, rule, and access restriction models. However, they fail to achieve higher security performance, which degrades the trust of data owners. To handle this issue, an efficient Adaptive Feature Centric Polynomial (AFCP) data security model is described here. The proposed method can be adapted to enforce security on any kind of data. The AFCP scheme classifies the features of EHR data under different categories based on their importance in being identified from the data taxonomy. By maintaining different categories of data encryption schemes and keys, the model selects a specific key for a unique feature with the use of the polynomial function. The method is designed to choose a dynamic polynomial function in the form of m(x) n, where the values of m and n are selected in a dynamic way. The method generates a blockchain according to the feature values and adapts the cipher text generated by applying a polynomial function to data encryption. The same has been reversed to produce the original EHR data by reversing the operation. The method enforces the Healthy Trust Access Restriction scheme in restricting malicious access. By adapting the AFCP model, the security performance is improved by up to 98%, and access restriction performance is improved by up to 97%. The proposed method increases the access restriction performance in the ratio of 19%, 16%, and 11% to HCA-ECC, EHRCHAIN, and PCH methods. Similarly, security performance is increased by 17% 13%, and 11% to HCA-ECC, EHRCHAIN, and PCH methods.
Read MoreDoi: https://doi.org/10.54216/JCIM.150103
Vol. 15 Issue. 1 PP. 22-33, (2025)
A polymorphic worm is a kind of worm that can change its payload in every infection attempt, so it can evade the Intrusion Detection Systems (IDSs) and perform illegal activities that lead to high losses. These worms can mutate as they spread across the network, causing most of the existing IDSs to carry out the polymorphic worm’s detection with high levels of both false positives and false negatives. In this paper, we propose a double-honeynet system that can detect polymorphic worm instances automatically. The Double-honeynet system is a hybrid system with both Network-based and Host-based mechanisms. This allows us to collect polymorphic worm instances at the network-level and host-level, which reduces the false positives and false negatives dramatically. The experimental deployment of a Double-honeynet network over a seven-day period successfully collected instances of various polymorphic worms, including 3511 Allaple, 3228 Conficker, 2817 Blaster, and 2452 Sasser worms. By utilizing, the Honeywall's Walleye interface; we were able to analyze the data and simulate the detection of these worms by generating new signatures, which were not previously recorded, demonstrating the system's capability to detect zero-day polymorphic threats. Analysis of Blaster worm instances revealed significant similarities in their payloads due to exe headers, indicating the necessity of preprocessing to remove these headers before signature generation, although the generation of signatures is beyond the scope of this study.
Read MoreDoi: https://doi.org/10.54216/JCIM.150104
Vol. 15 Issue. 1 PP. 34-49, (2025)