Journal of Data Mining and Knowledge Engineering ISSN: 3139-5341 (Online)

Abstract

Sequential pattern mining is an important model in data mining. Its mining algorithms discover all item sets in the data that satisfy the user-specified minimum support (minsup) and minimum confidence (mincon) constraints. Minsup controls the minimum number of data cases that a rule must cover. Mincon controls the analytical strength of the rule. Since only one minsup is used for the whole database, the model completely assumes that all items in the data are of the same nature and have similar frequencies in the data. In many applications, some data items appear frequently in the data, while others rarely appeared. If minsup is set too high, those rules that involve rare data items will not be found. To find rules that involve both frequent and rare items, minsup has to be set very low. This may affect combinational explosion because those frequent items will be associated with one another in all possible ways. This problem is called the rare item problem. This paper proposes to solve this problem. The technique allows the user to specify multiple minimum supports (MMS) to reflect the natures of the items and their mixed frequencies in the database. In data mining, different rules may need to satisfy different minimum supports depending on what items are in the database. Experiment results show that the technique is very effective. 

Keywords: Data Mining, Mining Sequential Patterns, Large Sequence Databases,

Abstract

The wide set of architectural design platforms available to corporate data centres has increased drastically over the years, which are spread across traditional proprietary data centres to multi-tenant structures. This model advocates a cloud based approach which can well be managed dynamically among internal and cloud based data centres. The requirement of this project is to combine all large file handling services under single code base, upgrade components into latest technology and leverage public cloud features. The plan is to deploy public facing gateway servers which collect files and persist it in Azure Blob. The latest data reported by IT teams put into lime light the concerns and issues faced by traditional security models which struggle to keep up with the emerging challenges and suggest that they are no more adequate for the challenges they face in addressing dynamic nature of cloud based Data centre This proposed dynamic cloud model provides many advantages which includes retrospective detection, dynamic management, effective deployment of huge files, improved detection rates and being cost effective

Keywords: Data centre, Azure Blob, Meta-data, Payload, Request Headers, File gate ways, Cassandra, Spring Boot web service

Abstract

Data Stream Mining is the process of extracting knowledge structures from continuous, rapid data records. Now a day’s huge amount of data is processed & analyzed. So it is very important to classify data & information properly. The information is basically unstructured & continuous. So huge volume of continuous data which has multidimensional feature & often fast changing. It is required to construct model which adapt such changes & give fast response. Such information flow examples are network traffic, sensor data, call center records etc. Class evolution is now a day’s important topic in data stream mining which handles such data. So in previous work proposed a model Class Based ensemble for Class evolution (CBCE) to maintain such a large amount of streams. But for complex & massive data result would be different. so complex class ensemble model (CCEM) is proposed for classification so huge & complex classes can be handled & classify & also proposed a model for class disappearance only so that more emp`hasize on class disappearance than class reoccurrence & novel class.

Keywords: Data stream mining, class evolution, ensemble model, incremental learning. Page 85-98 © MANTECH PUBLICATIONS 2017

Abstract

The proliferation of multimodal data in recent years—encompassing text, images, and sensor outputs—has presented unique opportunities and challenges in knowledge extraction. Traditional single-modal approaches often fail to capture the rich semantic relationships inherent across modalities. This review paper explores the state-of-the-art techniques, models, and frameworks for extracting knowledge from multimodal datasets. The paper discusses representation learning, fusion strategies, attention mechanisms, and domain adaptation, highlighting the role of deep learning architectures such as transformers and graph neural networks. Practical applications in healthcare, autonomous systems, smart cities, and industrial monitoring are examined. Challenges such as data heterogeneity, alignment, missing modalities, and interpretability are addressed. Future directions emphasize robust multimodal embedding, real-time knowledge extraction, and explainable models. The insights provided herein aim to guide researchers and practitioners in leveraging multimodal data for comprehensive decision-making and knowledge discovery.

Keywords: Multimodal Data, Knowledge Extraction, Deep Learning, Representation Learning, Fusion Techniques, Text-Image-Sensor Data, Graph Neural Networks, Attention Mechanisms

Abstract

Traditional data mining techniques mostly rely on pairwise relationships represented by graphs or relational tables. However, many real-world systems naturally contain interactions among more than two entities simultaneously, such as co-authorship networks, biological pathways, and group communications. These complex multi-entity relations can be better modeled using hypergraphs and higher-order structures. Hypergraph and higher-order data mining has emerged as an important research direction to capture group level dependencies, collective patterns, and multi-way associations that cannot be discovered using conventional graph-based methods. This paper presents a comprehensive review of hypergraph and higher-order data mining. We first introduce the theoretical foundations of hypergraphs and their properties. Then we discuss hypergraph representation learning, clustering, classification, and community detection methods. We also examine higher-order network mining including simplicial complexes and tensor-based approaches. Applications in social networks, recommender systems, bioinformatics, and knowledge graphs are analyzed. A comparative analysis of algorithms and challenges is also presented. Finally, open research directions such as scalability, interpretability, and dynamic hypergraphs are discussed. The study shows that hypergraph-based mining provides richer structural understanding compared to traditional graph mining and is essential for modeling modern complex data systems.

Keywords: Hypergraph mining, higher-order networks, multi-relational data, hypergraph clustering, tensor mining, group interactions

Abstract

CIBIL is India’s first Credit Information Company founded in August 2000. CIBIL collects and maintains records of an individual’s payments pertaining to loans and credit cards. These records are submitted to CIBIL by member banks and credit institutions, on a monthly basis. This information is then used to create Credit Information Reports (CIR) and credit scores which are provided to the credit institutions in order to help evaluate and approve loan applications. The existing CIBIL System does not provide any method to calculate score for the new users, whereas, this research is based on the methods to calculate score for both the existing as well as the new users.

Keywords: CIBIL, CIBIL TransUnion Score, Banks, Credit Information Report

Abstract

Cloud Computing is the Technology which is rapidly been used nowadays. This Technology is used for Hosting and Delivering Services over Internet. Use of Cloud Computing Minimizes the Efforts of Management in many fields. Increase in Use of Cloud Computing may results in some serious issues like Energy Consumption, power shortage. As per the growth in use the need of more servers is also increasing which causes more consumption of energy. Cloud Computing technology has data centres which runs with numerous hosts results in High Energy consumption. To Redice the Energy Consumption we are using Green Algorithm. It is an Eco-friendly way to use Computers and its resources. The Main aim is to study various techniques which are energy efficient to achieve Green Cloud Computing.

Keywords: Cloud Computing, Data Centres, Energy Consumption, Energy Efficiency Control, Green Cloud Computing.

Abstract

Web page can be divided in various blocks called as fragments. A fragment is a portion of a web page which has a distinct theme or functionality and is distinguishable from the other parts of the page. Dividing web pages into fragments has provided significant benefits. Good methods are needed for dividing web pages into fragments. Manual fragmentation of web pages is expensive, error prone, and un-scalable. Due to these problems, extraction of web fragments using Content extractor algorithm and DeSeA algorithm have been widely used. The proposed work has following features: 1) Detect fragment using content extractor algorithm. 2) Extraction of fragment detected in step (1). 3) Detect fragment using DeSeA algorithm. 4) Extraction of fragment detected in step (3). 5) Analyze results of extracted fragment using above algorithms.

Keywords: Fragment, Content Extractor, DeSeA.

Abstract

Relational pattern mining aims to discover meaningful structures, dependencies, and patterns from interconnected data such as social networks, biological systems, knowledge graphs, and recommendation platforms. Traditional relational learning and graph mining techniques often struggle with scalability, feature engineering, and capturing complex high-order dependencies. Graph Neural Networks (GNNs) have emerged as a powerful deep learning paradigm that directly operates on graph-structured data, enabling automatic learning of relational representations. This paper provides a comprehensive review of Graph Neural Networks for relational pattern mining. It discusses fundamental GNN architectures, relational representation learning mechanisms, pattern mining approaches, and applications across domains. The study also compares GNN-based relational mining with classical graph mining techniques and highlights challenges such as interpretability, scalability, and dynamic graph modeling. Finally, future research directions including explainable relational mining, temporal GNNs, and hybrid symbolic neural models are outlined. The paper aims to serve as a structured reference for researchers exploring relational knowledge discovery using graph deep learning methods.

Keywords: Graph neural networks, relational pattern mining, graph representation learning, knowledge graphs, link prediction, graph mining, deep learning on graphs

Abstract

The rapid growth of data-driven applications has significantly transformed decision-making processes across domains such as finance, healthcare, recruitment, and law enforcement. While data mining algorithms provide powerful tools to extract patterns and knowledge, they can inadvertently propagate or amplify societal biases present in the data. Fairness-aware data mining (FADM) algorithms have emerged as a solution to mitigate algorithmic discrimination and ensure equitable outcomes. This paper presents a comprehensive review of fairness-aware data mining techniques, focusing on pre-processing, in-processing, and post-processing strategies. We discuss fairness metrics, challenges in practical deployment, and provide comparative analysis of recent algorithms. Future research directions are also identified to promote fairness without compromising model utility.

Keywords: Fairness, Bias Mitigation, Data Mining, Algorithmic Discrimination, Fairness Metrics, Machine Learning

Abstract

Knowledge graphs (KGs) have become a foundational tool in artificial intelligence, supporting applications in recommendation systems, natural language understanding, and decision-making. While reasoning over knowledge graphs enhances the ability to infer implicit knowledge, many methods remain opaque, leading to limited trust and interpretability. Explainable reasoning over knowledge graphs aims to make the inference process transparent, interpretable, and understandable by humans. This review provides an overview of current methods for explainable reasoning, including rule-based, embedding-based, and hybrid approaches. We discuss evaluation metrics, datasets, and challenges in the field. Moreover, we highlight emerging trends such as neuro-symbolic reasoning and graph attention mechanisms. Through comparative analysis, this paper presents insights into the trade-offs between reasoning performance and explainability, emphasizing the need for methods that balance accuracy with interpretability.

Keywords: Knowledge Graphs, Explainable AI, Reasoning, Interpretability, Graph Embeddings, Rule-based Reasoning

Abstract

Pattern mining and machine learning models are widely used in diverse domains such as healthcare, finance, and cybersecurity to extract meaningful insights from complex datasets. However, the increasing complexity of models has raised concerns about their interpretability and transparency. Explainable pattern mining (XPM) and interpretable models provide methods to not only uncover patterns but also make the results understandable for humans. This review paper explores the current research trends in explainable pattern mining, examines different interpretable modeling techniques, and discusses their applications, benefits, and limitations. The paper also highlights challenges and future research directions in building models that balance predictive performance with explainability.

Keywords: Explainable AI, Pattern Mining, Interpretable Models, Transparent Machine Learning, Knowledge Discovery, Model Explainability

Abstract

Large-scale heterogeneous networks are widely present in many real world systems such as social media platforms, citation graphs, biological interaction networks and e-commerce ecosystems. Unlike homogeneous networks where all nodes and edges are of same type, heterogeneous networks consist of multiple node and relation types, which makes community detection more challenging and also more meaningful. Community detection aims to identify groups of nodes that are densely connected or semantically related. In heterogeneous environments, communities often represent functional modules, user interest groups or knowledge clusters. This paper reviews the recent progress in community detection methods for large-scale heterogeneous networks. We discuss network characteristics, formal problem definitions, classical and modern detection approaches including meta-path based methods, probabilistic models, embedding techniques and deep learning approaches. Scalability challenges and evaluation metrics are also analyzed. Comparative discussion shows that embedding and graph neural network models are promising for large-scale settings but still face interpretability and efficiency issues. The paper finally outlines open research directions such as dynamic heterogeneous communities and privacy-aware detection.

Keywords: heterogeneous networks, community detection, graph mining, network embedding, meta-path, large-scale graphs

Abstract

Knowledge bases (KBs) have become fundamental resources for artificial intelligence, natural language processing, and data-driven decision-making. However, KBs are prone to various forms of bias that can negatively influence downstream applications, leading to unfair or discriminatory outcomes. Bias in KBs may arise due to data imbalance, human annotation errors, or skewed representation of entities and relationships. This paper presents a comprehensive review of existing methods for bias detection and mitigation in knowledge bases. We categorize techniques into data-driven, embedding-based, and algorithmic approaches, discussing their advantages and limitations. We also highlight benchmark datasets used for evaluating bias and propose a framework for integrating bias mitigation in KB construction. Finally, we discuss open challenges and future research directions.

Keywords: Knowledge bases, Bias detection, Bias mitigation, Fairness, Embedding, Knowledge representation

Abstract

The increasing volume and complexity of data in modern enterprises demand automated approaches to extract actionable insights efficiently. Autonomous data mining pipelines, combining Automated Machine Learning (AutoML) and Automated Knowledge Discovery in Databases (AutoKDD), represent a paradigm shift in handling end-to-end data workflows with minimal human intervention. This paper reviews the state-of-the-art methodologies in AutoML and AutoKDD, highlights their integration into autonomous pipelines, and evaluates their capabilities, limitations, and applications across industries. Key challenges such as model interpretability, scalability, and ethical considerations are discussed. The paper also provides a comparative analysis of prominent tools and frameworks, emphasizing best practices for pipeline design and deployment.

Keywords: AutoML, AutoKDD, Autonomous Data Mining, Machine Learning Pipelines, Knowledge Discovery, Data Analytics, Automated Model Selection

Abstract

Automated knowledge discovery has become a critical task in modern digital environments where massive textual and multimodal data are generated continuously. Traditional data mining and knowledge discovery approaches rely heavily on structured databases and manual feature engineering, which limits scalability and semantic understanding. Large language models (LLMs) provide a new paradigm for extracting knowledge directly from unstructured sources such as scientific literature, technical reports, social media, and enterprise documents. These models demonstrate strong capabilities in representation learning, semantic reasoning, summarization, and relationship extraction, enabling automated discovery of concepts, patterns, and insights. This paper reviews the role of large language models in automated knowledge discovery, covering architectures, methodologies, application domains, and evaluation strategies. We discuss how LLM-based pipelines perform knowledge extraction, entity discovery, relation mapping, and hypothesis generation. Advantages such as contextual awareness, cross-domain generalization, and interactive learning are examined along with limitations including hallucination, bias, and explainability challenges. A conceptual framework for LLM-driven knowledge discovery is presented along with comparative analysis of classical and LLM-based methods. The review highlights emerging research directions including autonomous scientific discovery, domain-adapted LLMs, hybrid symbolic-neural reasoning, and trustworthy knowledge systems. Automated knowledge discovery using LLMs is expected to transform scientific research, engineering analytics, healthcare intelligence, and industrial decision support.

Keywords: Large language models, knowledge discovery, automated reasoning, text mining, semantic analysis, knowledge graphs

Abstract

As the knowledge available on the World Wide Web is increasing every day, many users find it difficult to get the data they need as they don't use any recommendation system to scan the web. A recommendation system can significantly reduce the time taken for finding the data on the web. However, even some recommendation systems don't give the precise information users need as these systems don't take into account the user's interests. As analyzing user interests and identifying the right domain is essential to in Web Usage Mining. Detecting user's interests is possibly by scanning their Web Log Data, and the information obtained from there will be useful for website developers and users alike. In this paper, a unique algorithm is presented that can extract users' interests accurately as it is based on visit density and time. This method is very useful in finding a user's interested domains.

Keywords: Web Mining, Web Usage Mining, Data Mining, Weblog Data, Web Content Mining

Abstract

Data mining is widely used in healthcare applications. However, data in healthcare are continuous in nature and data mining algorithms cannot directly deal with it. Discretization of continuous attributes improves the efficiency of data mining algorithms. This paper compares the performance of different discretization techniques on three different University of California, Irvine (UCI) datasets namely Wisconsin Diagnostic Breast Cancer (WDBC), Breast tissue and PIMA Indian diabetes in terms of classification accuracy. The aim of this study is to identify the discretization technique that helps to improve the accuracy of the classification process. The experimental results have showed that the entropy based discretization gives the highest classification accuracy using Naïve Bayes classifier.

Keywords: Breast Cancer, Breast Tissue, Indian Diabetes, Classification, Discretization