In today’s digitally connected world, where cyber threats are becoming increasingly complex, finding modern and secure text encryption solutions that maintain maximum runtime performance while offering high-level protection is more crucial. The deployment of sophisticated security paradigms is often accompanied by a significant escalation in computational overhead. Thus, the fundamental objective resides in the mitigation of computational overhead while maintaining an uncompromising security posture. Internet of Things (IoT) devices require strong security measures for data transmission. Also, protecting communication channels against illegal access and eavesdropping has become crucial due to the exponential expansion of the IoT. The IoT implementations frequently have weak, unencrypted data streams that are susceptible to manipulation and interception. In order to overcome this, the proposed work incorporates lightweight protection using Moving Picture Experts Group (MPEG) derived motion vectors and dual encryption techniques to guarantee message confidentiality and integrity via limited IoT networks. The proposed method starts with resizing MPEG video frames to dimensions [1080, 1920]. After extracting motion vectors from two successive video frames, scale the obtained vectors to 1000. The exclusive OR (XOR) procedure is applied to the combined motion vectors. A one-dimensional (1D) vector is then produced. The initial elliptic curve Diffie-Hellman (ECDH) private key is created using a mapping of a hash function. The public keys, shared secret keys, and a second private key are also created. The shared secret key is used to generate the Advanced Encryption Standard (AES) main key. After that, the created AES is used to encrypt and decrypt text messages ranging in length from 10 to 300 bytes. Several evaluation metrics, including mean square error (MSE), peak signal to noise ratio (PSNR), correlation coefficient (CC), avalanche Effect (AE), and compression ratio (CR) values, are evaluated between the original and ciphertext. The presented method has demonstrated optimal performance in terms of encryption and decryption times as well as public and private key generation. Thus, improving the IoT application’s overall security condition by guaranteeing that only authorized endpoints can decrypt and read the data, and showing minimal latency overhead as compared to insecure transmission. This suggests that it is a highly effective solution for secure text communication, offering lightweight encryption suitable for a wide range of resource-constrained and real-time applications.

The widespread use of social media has made assessing users’ tastes and preferences increasingly complex and important. At the same time, the rapid dissemination of misinformation on these platforms poses a critical challenge, driving significant efforts to develop effective detection methods. This study offers a comprehensive analysis leveraging advanced Machine Learning (ML) techniques to classify news articles as fake or true, contributing to discourse on media integrity and combating misinformation. The suggested method employed a diverse dataset encompassing a wide range of topics. The method evaluates the performance of five ML models: Artificial Neural Networks (ANNs), Convolutional Neural Networks (CNNs), Long Short-Term Memory networks (LSTMs), Decision Trees (DTs), and Support Vector Machines with Radial Basis Function (SVM-RBF) kernels. The presented methodology included thorough data preprocessing, detailed parameter tuning during model training, and robust statistical analyses to ensure fair and accurate performance comparisons. The results demonstrate that the combination of Term Frequency-Inverse Document Frequency (TF-IDF) with ANN and CNN achieved the highest accuracy of 99.13%, showcasing the effectiveness of these approaches in text-based news classification. The LSTM model followed closely with an accuracy of 98.59%, while the DT and SVM-RBF models achieved accuracies of 85.67% and 90.22%, respectively. These findings highlight the superior performance of deep learning (DL) models when combined with effective feature extraction techniques such as TF-IDF. The models offer practical utility and show promising potential for integration into editorial workflows to facilitate pre-publication news verification. Furthermore, statistical test methods such as Analysis of Variance (ANOVA) and Tukey’s Honestly Significant Difference (HSD) tests are also performed. The obtained results clarify significant performance differences among the evaluated models, highlighting their unique capabilities and comparative strengths in the context of fake news detection. Hence, the presented study reinforces the importance of artificial intelligence based tools in promoting media reliability and provides a foundation for future advancements in automated misinformation detection systems.

Ensuring the secure transmission of secret messages, particularly through video—one of the most widely used media formats—is a critical challenge in the field of information security. Relying on a single-layered security approach is often insufficient for safeguarding sensitive data. This study proposes a triple-lightweight cryptographic and steganographic model that integrates the Hill Cipher Technique (HCT), Rotation Left Digits (RLD), and Discrete Wavelet Transform (DWT) to embed secret messages within video frames securely. The approach begins with encrypting the secret text using a private key matrix (PK₁) of size 2 × 2 up to 6 × 6 via HCT. A second encryption layer is applied using a dynamic private key (PK₂) derived from the RGB pixel values of the video frame, resulting in a rotated cipher. The doubly encrypted message is then embedded into the video frames using the DWT method. Upon transmission, the concealed message is extracted using inverse DWT and decrypted in two steps—first with PK₂ and then with the inverse of PK₁. Experiments conducted using MPEG video sequences and message lengths ranging from 10 to 300 bytes demonstrate strong performance in terms of Mean Square Error (MSE), Peak Signal-to-Noise Ratio (PSNR), and Correlation Coefficient (CC) between original and encrypted messages. The similarity between original and stego frames is further validated using Structural Similarity Index (SSIM), Mean Absolute Error (MAE), Number of Pixel Change Rate (NPCR), and Unified Average Changing Intensity (UACI). Results confirm that utilizing video frames to generate PK₂ offers superior security compared to static key images. Moreover, the indistinguishability between original and stego frames highlights the method’s robustness against visual and statistical attacks.