Software vulnerabilities pose significant risks to the security and reliability of modern systems, making automated vulnerability detection an essential research area. Traditional static and rule-based approaches are limited in scalability and adaptability, motivating the adoption of data-driven methods. In this survey, we present a comprehensive review of Machine Learning (ML), Deep Learning (DL), and Large Language Models (LLMs) techniques for vulnerability detection. We analyze recent advances in feature representation, fine-tuning strategies, generative approaches, and prompt engineering, while highlighting their ability to capture both syntactic and semantic properties of source code. Furthermore, we examine commonly used evaluation metrics and provide a critical discussion of key challenges, including the lack of large-scale real-world datasets, limited vulnerability coverage, class imbalance, interpretability gaps, hallucination, and high computational costs. To address these issues, we outline promising future research directions, such as neuro-symbolic hybrid methods, parameter-efficient fine-tuning, continual learning, cross-language generalization, and explainable AI for vulnerability detection. Unlike previous studies, the present work explores learning paradigms from ML to LLMs using comprehensive evaluation criteria that highlight analytical capability, feature interpretability, and code-context comprehension. By combining these factors, our study addresses the methodological gap between classic feature-based approaches and current LLM-driven reasoning frameworks, providing beneficial insights to develop robust, scalable, and trustworthy software vulnerability detection systems.

Outlier reduction is crucial in computer science for improving data quality, analysis accuracy, and modeling robustness. Selection and modification of DBSCAN parameters are essential for optimal clustering accuracy and outlier detection. We developed an adaptive technique to minimize outliers in the DBSCAN algorithm using a linear congruential method (LCM) to determine values of Epsilon (Eps) and Min-Points (MinPts), known as modified DBSCAN (M-DBSCAN). To enhance the DBSCAN method, we create integer random numbers for MinPts (1--100) and floating numbers for Eps (0.1--1.5) using LCM. We adjusted parameter lists to reduce outliers based on MinPts and Eps values. We choose parameters based on dataset features and requirements, balancing clustering sensitivity and noise treatment. For experiment result analysis we use the Silhouette Score (SS) method. M-DBSCAN improved all cases and it has 50% poorer outlier accuracy than DBSCAN.

Bitter peptides are short amino acid chains that produce a bitter taste. These peptides are made primarily in food processing through the chemical reduction of peptides. The bitterness arises from the specific sequence of amino acids in peptides, which interact with the bitter taste receptors on the human tongue. These peptides influence nutrition and health, offering insights into protein digestion and bioactive advantages. Hence, correctly identifying bitter peptides is pivotal for revealing the biochemical properties of efficient medication. The computational approach is most suitable for identifying bitterness, where most studies obtained insufficient outcomes. Therefore, the current study developed an ensemble-based framework called "BitterEN", where we integrate the Gradient Boosting (GB) and Multi-layer Perception (MLP) methods. Our proposed method improved more than 3 % of accuracy compare to all of the state-of-the-arts methods, where the proposed approach achieved 0.995 accuracy in merged feature extractions with the Random Forest (RF) feature selection method. We used 50 iterations over the performance evaluation phases to enable a more exact generalization of model performance. In addition, we provided a convenient GitHub-based version of our bitter peptide identification. It highlights the practical applicability of these findings. We are optimistic that the proposed approach might benefit many fields, including healthcare development and nutritional science.

The plague of cyberbullying on social media exerts a dangerous influence on human lives. Due to the fact that online social networks continue to daily expand, the proliferation of hate speech is also growing. Consequentially, distressing content is often implicated in the onset of depression and suicide-related behaviors. In this paper, we propose an innovative framework, named as the trustable LSTM-autoencoder network (TLA NET), which is designed for the detection of cyberbullying on social media by employing synthetic data. We introduce a state-of-the-art method for the automatic production of translated data, which are aimed at tackling data availability issues. Several languages, including Hindi and Bangla, continue to face research limitations due to the absence of adequate datasets. Experimental identification of aggressive comments is carried out via datasets in Hindi, Bangla, and English. By employing TLA NET and traditional models, such as long short-term memory (LSTM), bidirectional long short-term memory (BiLSTM), the LSTM-autoencoder, Word2vec, bidirectional encoder representations from transformers (BERT), and the Generative Pre-trained Transformer 2 (GPT-2), we perform the experimental identification of aggressive comments in datasets in Hindi, Bangla, and English. In addition to this, we employ evaluation metrics that include the F1-score, accuracy, precision, and recall, to assess the performance of the models. Our model demonstrates outstanding performance across all the datasets by achieving a remarkable 99% accuracy and positioning itself as a frontrunner when compared to previous works that make use of the dataset featured in this research.