Journal of IoT Security and Smart Technologies (e-ISSN: 2583-6226)

AI-driven Intrusion Detection System Using Machine Learning: An Extensive Study

Rashmi — 2026-09-09

Network Intrusion Detection Systems (NIDS) have gained significant importance in modern cybersecurity owing to the escalating frequency and complexity of cyber-attacks on enterprise networks, cloud computing platforms, and Internet of Things (IoT). Traditional signature-based NIDS cannot identify zero-day attacks and evolving attacks, which is why intelligent techniques like machine learning (ML) and deep learning have been introduced to identify threats. This survey reviews recent advancements in machine-learning-based and deep-learning-based intrusion detection by comparatively analyzing five primary research studies and supporting literature covering different datasets, attack scenarios, learning approaches, and research challenges. In this survey, common datasets used, feature engineering techniques, classifiers, and metrics have been described along with their pros and cons. Moreover, key challenges faced by current NIDS include dataset imbalance, false positives, inadequate cross-dataset generalization, detection of unknown attacks, computational cost, and lack of interpretability. Additionally, potential future research directions for intrusion detection, namely explainable artificial intelligence (XAI), federated learning, transformer-based IDSs, continual learning, and adaptive hybrid models, have been discussed in this survey.

Lightweight Authentication and Encryption Protocols for Resource-constrained IoT Devices: A Review

Sonam Ashokrao Rewatkar — 2026-09-14

The rapid proliferation of Internet of Things (IoT) devices across healthcare, industrial automation, smart cities, and consumer applications has intensified the need for security mechanisms that operate within tight constraints on memory, computation, and energy. Conventional cryptographic primitives such as AES-256 and RSA, while secure, are frequently unsuitable for microcontrollers, RFID tags, and sensor nodes with kilobyte-scale memory and battery-limited operation. This has driven a distinct research area — lightweight cryptography — dedicated to authentication and encryption schemes tailored to resource-constrained hardware. This review synthesizes recent literature (2014–2025) on lightweight authentication and encryption protocols for IoT, organizing the field into five strands: lightweight block ciphers and authenticated encryption, including the NIST-standardized ASCON family; elliptic-curve-based mutual authentication; physically unclonable function-based schemes; RFID-specific ultra-lightweight protocols; and blockchain-assisted authentication architectures. Across these strands, the review compares reported computational overhead, memory footprint, energy consumption, and resilience against common IoT threats such as impersonation, desynchronization, and modeling attacks. The analysis shows that no single approach dominates across all constraint profiles: block ciphers suit bulk data confidentiality, PUFs suit tamper-evident device identity, and ECC suits scenarios requiring public-key functionality at higher, though still bounded, cost. The review concludes by highlighting persistent gaps, including post-quantum readiness and the recurring cryptanalysis-and-redesign cycle in ultra-lightweight RFID schemes, and proposes directions for future standardization and formal verification efforts.

A Deep Learning Framework for Intelligent Network Intrusion Detection and Threat Classification

Shraddha Solanki — 2026-10-03

In many areas of application, making better decisions and using predictive analytics requires correctly classifying large, complicated datasets. Unfortunately, problems like missing values, class imbalance, redundant features, and high-dimensional data make deep learning models less useful. For better classification performance, this study suggests a complete deep learning-based system that combines advanced data preparation, dimensionality reduction, feature engineering, and residual neural network designs. Firstly, the collected dataset is preprocessed by handling missing values, encoding labels, and using SMOTE to fix class imbalance and improve data quality. Following this, Principal Component Analysis (PCA) is used to reduce the dimensionality while preserving the most valuable features. The discriminative power of the dataset is improved using feature engineering methods such as feature selection and extraction. Statistical and pattern analysis are then used to understand the basic qualities of the data. Later, the cleaned dataset is utilized to teach several deep learning models, such as ResNet34, ResNet50, and ResNet101. The optimum model for classification is determined by comparative evaluation, and the final predictions are made using the best-performing design. Results like AUC-ROC, F1-score, recall, precision, and accuracy in classification are expected to improve using the suggested framework, which is also expected to decrease processing complexity and overfitting. A quick, scalable, and dependable classification framework that can handle imbalanced and high-dimensional datasets in a variety of real-world scenarios is produced by combining robust preprocessing, enhanced feature representation, and deep residual learning.

Digital Evidence Timeline Builder: A Browser-Based Tool for Digital Forensic Timeline Analysis

Dharani Dharan M. — 2026-10-05

Organizing digital evidence into a structured chronological view is a routine but labour-intensive task in digital forensics and cyber incident response. Analysts often do this manually in spreadsheets or documents, which offer little support for structured metadata, filtering, or visual review. This paper presents the Digital Evidence Timeline Builder, a prototype browser-based tool implemented in HTML, CSS, and JavaScript. It allows an analyst to manually record evidence events (timestamp, category, severity, description, artefact source, file hash, analyst name, and tags), sorts them chronologically, displays them on an interactive timeline grouped by date, filters them by category and severity, summarizes them on a dashboard, exchanges case data as JSON, and generates a PDF report. The tool is not an artefact parser or an acquisition/analysis suite. It targets a niche that established tools such as Plaso, Timesketch, Autopsy, and DFIR-IRIS serve differently: a zero-installation, offline, analyst-curated timeline for evidence that has already been acquired. The prototype was evaluated by functional testing against a documented protocol comprising defined test cases, explicit pass/fail criteria, a described test environment, and a repeatability procedure. Results are reported as qualitative functional verification (pass/fail); no accuracy percentages and no measured efficiency gains are claimed. The paper also states the prototype’s forensic-validity limitations: it does not validate timestamps, handle time zones or clock drift, compute or verify hashes, enforce chain of custody, keep an audit log, or protect exported case files from modification. The tool is therefore a documentation and visualization aid to be used alongside, not instead of, forensically sound acquisition, preservation, and analysis procedures.

HealthChain: A Privacy-Preserving and Verifiable Healthcare Record Management System Using Permissioned Blockchain

Dinesh Jadhav — 2026-10-17

The digitization of healthcare records has introduced significant challenges regarding data privacy, security, and interoperability. Centralized Electronic Health Record (EHR) systems are prone to single points of failure, insider manipulation, and unauthorized access, leaving patients with negligible control over their sensitive medical information. This article proposes HealthChain, a decentralized healthcare record management system built on a permissioned blockchain (Hyperledger Fabric). HealthChain utilizes a dual-tier storage strategy where sensitive Protected Health Information (PHI) is stored off-chain in an encrypted PostgreSQL database, while only cryptographic hashes and provenance metadata are anchored to the immutable ledger. The study introduces a dynamic patient-controlled consent protocol that enables fine-grained access control, allowing patients to grant or revoke access to specific medical records for defined temporal windows. Furthermore, the system incorporates a mandatory human-in-the-loop verification process for healthcare providers to ensure clinical accountability before key issuance. Experimental performance evaluations demonstrate that HealthChain maintains low transaction latency, linear scalability under multi-user concurrency, and absolute mathematical verifiability against internal and external data tampering.