Efficient Identification of Cyber Threat using Artificial Neural Networks

Abstract Education is a critical driver of social inclusion and economic development. In India, despite constitutional safeguards and targeted government programmes, religious minorities continue to face disparities in educational access and outcomes. This study examines the barriers affecting the effective implementation of educational policies for religious minorities in India through a qualitative analysis of secondary data. Drawing on government reports, policy documents, census data, and peer-reviewed literature, the paper analyses structural, socio-economic, cultural, and administrative constraints that limit policy effectiveness. Regional case studies from North and South India highlight variations in implementation and outcomes. The findings reveal persistent gaps between policy design and on-ground delivery due to limited awareness, bureaucratic delays, funding discontinuities, and socio-cultural constraints. The paper concludes with policy-oriented recommendations to strengthen monitoring, enhance inclusivity, and improve educational outcomes for religious minority communities. Keywords: Minority Education, Educational Policy, Policy Implementation, Religious Minorities, India Corresponding Author: Savitha AC*, savithaac@jssateb.ac.in 1.Introduction The fast-passed digital information of the world wide system and the constant connectivity in today's society have completely changed the way information security works. While these connections promote innovation and economic progress, they also create many more opportunities for cyber criminals to attack. As a result,the number and complexity of cyber threats are steadily increasing from common attacks to more advanced ones. These threats range from malware and phishing campaigns to sophisticated, targeted attacks known as advanced persistent threats (APTs) that focus on critical national infrastructure, corporate espionage, and data theft. Old ways of keeping systems safe - built on known patterns and set-in-stone rules - are starting to fail more often. These models struggle when facing unknown threats that appear without warning. Because today’s networks handle so much information, coming fast in many shapes, old tools get buried under the load. Staying ahead means building defenses that adjust on their own, move quickly, and clearly spot harm hiding inside everyday actions while it happens. Now driving much of today’s defence planning is a growing reliance on artificial intelligence and machine learning. Since deep learning handles messy inputs - traffic snapshots, system records, actions people take - it spots subtle attack clues others simply overlook. Artificial neural networks step in, letting companies spot risks sooner - often ahead of major harm. ANN designs tackle cyber security challenges across the digital world.are: Feed-forward Neural Networks (FNNs): They are used for basic classification tasks, helping to sort network traffic into groups like normal activity, DoS attacks, or probing attempts. Convolution Neural Networks (CNNs): It is mainly used for image processing, they are also adapted to detect patterns and structures in organised data—like analysing byte sequences in malicious files or entries in system logs. Recurrent Neural Networks (RNNs) and Long Short-Term Memory (LSTM) networks: They are important for analysing how network traffic changes over time. These models can remember previous events in a session, which helps in detecting multi-stage attacks and unusual activities that develop gradually. 2.Literature Survey Neural computing is becoming a key part of modern cyber security. Today’s cyber attacks are fast, complex, and constantly changing, so we need smarter methods to detect them. Many research papers especially from IEEE and Google Scholar show that Artificial Neural Networks (ANNs) are helping us. Different ANN models work in different ways- CNNs help find patterns in data packets and malware files, while RNNs and LSTMs are good at understanding how network traffic changes over time, which helps catch multi-step attacks. Studies by researchers 1,2 show that ANN based systems often perform better than older methods like SVMs or Decision Trees on datasets such as NSLKDD and CICIDS2017. Other researchers 3,4 have shown how ANNs can protect IoT devices and even analyse text from OSINT and dark web sources to predict future threats. Overall, research 5-8 clearly shows that neural networks make threat detection faster, smarter, and more accurate. But they also have challenges: they need a lot of good-quality training data, they are hard for humans to interpret because of their “black box” behaviour, and they often require heavy computing power. Because of this, experts suggest using hybrid models that mix different ANN types with Explainable AI (XAI) so systems can be accurate, scalable, and understandable. This combination is expected to build stronger and more reliable cyber security systems for the future. When hackers get smarter, old ways of spotting break-ins start to fail. Systems using just known patterns or simple number checks can miss sneaky threats. A fresh approach has been gaining attention lately. Instead of relying on those outdated methods, it uses a special setup called a Directed Acyclic Graph 9,10. This framework links clues together in a smart flow. Information travels step by step across its branches. Each stage adds clarity instead of noise. The path leads straight to a clear conclusion about whether an attack happened. Confidence grows with every connected piece. Recent studies show how well this method holds up under pressure. Researchers 11 show that using both D(DAG) with a Belief Rule Base (BRB) to create a more organized, understandable, and accurate system for identifying security threats. In this approach, the DAG is used to map out the logical and layered relationships between different network features, attack categories, and detection steps, helping the system recognise the multi-level connections found in real network traffic. Each node in the DAG represents a specific security element, while the BRB layer handles uncertain, incomplete, or confusing data by using belief values and rule-based reasoning. Research 12 discussed the deep learning architectures like fully connected networks, stacked auto encoders, and hybrid deep models which consistently achieve better accuracy, stronger resilience, and improved generalization across a wide variety of datasets. 3.Implementation The proposed approach for a cyber threat detection system, built on the AI-SEM Artificial Intelligence Security. The model is trained using the back-propagation algorithm along with the Adam optimizer. Dropout, batch normalization, and early stopping techniques are applied to prevent over-fitting and help the system generalize better. After training finishes, performance gets checked using fresh network data. This shows accuracy of how the system responds, when it sees unfamiliar patterns performed. Confusion matrices show detailed outcomes in a grid format. Each of these plays a role in the full picture detection strength. The ANN picked up on subtle behavior trends. After everything has run its course, the results will be evident. Patterns started making sense after a while, quietly revealing connections underneath. Information from networks helps spot complex dangers like flooding systems or testing defenses Strange networks of infected machines, along with odd digital behaviors popping up. 4.System Overview Figure 1 shows the architecture and working of the proposed system. It has mainly three major units: (i) data processing unit, (ii) learning engine of ANN unit and (iii) real time threat detection unit. The first unit is responsible for providing input to the ANN and transferring the data with the preprocessing. Data aggregation and event profiling are done in the AI-SIEM system. The generated event data sets, vectors, and profiles are used for further processes. This phase works as protection for raw data. In the second stage prepossessed data are given to three artificial neural networks to find the best model for accuracy. In the real time threat detection phase , security analysis is done through ANN mechanical model. Fig.1: Proposed System Architecture The proposed methodology are given below: Load and preprocess the training dataset using TF-IDF. Generate event vectors for analysis. Profile and train the Neural Network model. Apply SVM, KNN, Naive Bayes, and Decision Tree algorithms. Compare performance using accuracy, precision, recall, and F1-score graphs. The system begins by loading a labelled network traffic data set containing both normal and malicious activities. This data set provides the features necessary for the ANN to learn attack patterns, classify intrusions, and evaluate detection accuracy.The TF-IDF algorithm processes raw network logs into numerical feature vectors by identifying the importance of each term. This highlights relevant keywords, reduces noise, and enables the model to distinguish between normal and malicious patterns. Each processed network record is transformed into a structured numerical representation called an event vector, which serves as the input for the ANN. ANN profiling examines how the model responds to different event vectors, learning the behavioural patterns of normal and malicious activities. The Support Vector Machine algorithm is applied to classify network activities by finding the optimal hyperplane separating normal and malicious traffic, improving detection. The k-Nearest Neighbors algorithm classifies new network events based on similarity to neighbouring data points, identifying threats through pattern comparison. Naive Bayes uses probabilistic calculations to classify network traffic based on the likelihood of features belonging to attack or normal classes. The Decision Tree algorithm splits