Machine Learning (ML) models are now a part of our everyday lives and significantly impact each individual. When users are given a clear and relevant explanation for a decision made by the model, they can understand the reasons and logic behind the model’s behavior, which helps to build trust in the model. This article proposes a zeroth level of explainability for an Optimal Random Forest model, which incorporates an Optimal Sparse Decision tree that is self-explainable. The methodology also involves vectorizing subtrees relevant to an instance, followed by dimensionality reduction and clustering. This method enhances explainability, ensuring high fidelity in the explanations provided. It also proposes a conversational interface for explaining the model’s decisions. Experimental analysis was done on four benchmark datasets, namely SMS Spam collection dataset (classification problem of ham or spam messages), Iris dataset (classification problem of Setosa or Non-Setosa based on length and width information of both sepal and petal data), Tic-tac-toe endgame dataset (based on board position content and their corresponding class labels), Heart Disease dataset (classification problem of prone to heart disease or not). Experimental analysis has demonstrated the explainability of the Optimal Random Forest model without any compromise on generalization and performance. Comparisons were also made with Shapley Additive Explanation (SHAP) derived from both a generalized random forest model and a Support Vector Machine model, and also on Rule learning + Voting model. The results have demonstrated the merits of the algorithm.

Abstract Topic modelling is a Natural Language Processing (NLP) technique that has gained popularity in the recent past. It identifies word co-occurrence patterns inside a document corpus to reveal hidden topics. Graph Neural Topic Model (GNTM) is a topic modelling technique that uses Graph Neural Networks (GNNs) to learn document representations effectively. It provides high-precision documents-topics and topics-words probability distributions. Such models find immense application in many sectors, including healthcare, financial services, and safety-critical systems like autonomous cars. This model is not explainable. As a matter of fact, the user cannot comprehend the underlying decision-making process. The paper introduces a technique to explain the documents-topics probability distributions output of GNTM. The explanation is achieved by building a local explainable model such as a probabilistic Naïve Bayes classifier. The experimental results using various benchmark NLP datasets show a fidelity of 88.39% between the predictions of GNTM and the local explainable model. This similarity implies that the proposed technique can effectively explain the documents-topics probability distribution output of GNTM.

Text clustering reveals inherent patterns from unlabelled datasets, a crucial capability for summarizing extensive text corpora via keyword extraction. This study presents a novel two-tier approach that utilizes string kernels and similarity measures to enhance spectral clustering algorithms for keyword extraction from text data. The first phase implements spectral clustering with precomputed affinity matrices derived from string kernels and similarity measures, evaluated using standard benchmarks. The second stage introduces a recursive spectral clustering technique for the hierarchical organization of the data, followed by the application of topic modeling to each cluster and sub-cluster for extracting semantically rich keywords. Tested on the BBC News, AG News, and Reuters datasets, the proposed methodology demonstrates a better representation of text data for downstream applications. The integrated approach synergizes spectral clustering, string kernels, hierarchical clustering, and topic modeling, establishing a robust and comprehensive framework for text data analysis and keyword extraction.

In the world of big data, extracting meaningful insights from large and continually growing distributed datasets is a major challenge. Classical clustering algorithms are effective at identifying clusters with convex structures. However, they fall short in identifying arbitrary-shaped clusters (more irregular and complex patterns), which are often encountered in real-world applications. The process of identifying non-convex cluster representations from very large and growing datasets is a challenge. It is further compounded by the distributed nature of the data, necessitating complex computations across multiple devices. Support Vector Clustering (SVC) is a much-celebrated algorithm capable of finding arbitrarily shaped clusters. However, the major limitation of this algorithm is that it will not scale to large volumes of data as the time and space complexity is high. The second limitation of the SVC algorithm is the requirement for large computation time in finding cluster structures. The adoption of a coreset based methodology is required for finding the true representation of the underlying large datasets. The implementation of hierarchical clustering on these distributed coresets, unlocks the potential to uncover a structured hierarchy of abstractions across the disseminated data. Moreover, a distance-based clustering approach guarantees the identification of clusters with diverse and arbitrary shapes, providing a robust framework for detecting complex structures. This research utilizes the Core Vector Machine (CVM) approach using an approximate Minimum Enclosing Ball (MEB) algorithm to efficiently address the complexities inherent in traditional SVC. Additionally, an enhanced medoid algorithm is employed for cluster head identification across the data sources. Hierarchical clustering is performed in the Reproducing Kernel Hilbert Space (RKHS) using cosine similarity distance matrices. This is used to identify compact non-convex clusters within distributed datasets. Performance assessment involves benchmarking our approach against state-of-the-art improved SVC algorithms using large datasets. The outcomes validate the superior performance of our approach compared to existing methods.

The success of machine learning in real-world use cases has increased its demand in mission-critical applications such as autonomous vehicles, healthcare and medical diagnosis, aviation and flight safety, natural disaster prediction, early warning systems, etc. Adaptive Boosting (AdaBoost) is an ensemble learning method that has gained much traction in such applications. Inherently being a non-interpretable model, the interpretability of the AdaBoost algorithm has been a research topic for many years. Furthermore, most of the research being conducted till now is aimed at explaining AdaBoost using perturbation-based techniques. The paper presents a technique to interpret the AdaBoost algorithm from a data perspective using deletion diagnostics and Cook’s distance. The technique achieves interpretability by detecting the most influential data instances and their impact on the feature importance of the model. This interpretability enables domain experts to accurately modify the significance of specific features in a trained AdaBoost model depending on the data instances. Unlike explaining AdaBoost using perturbation-based techniques, interpreting from a data perspective will enable it to debug data-related biases, errors and to impart the knowledge of the domain experts into the model through domain aware fine-tuning. Experimental studies were conducted with diverse real-world multi-feature datasets to demonstrate interpretability and knowledge integration through domain-aware fine-tuning.

Topic models are unsupervised machine learning techniques that output clusters of “topics” represented as co-occurring words with their associated probability distributions. Topic modeling algorithms find latent themes from large document collections by understanding their context. On the other hand, string kernels are supervised machine-learning techniques that quantify string similarities without explicit string encoding. We propose TopicStriKer, a model combining the advantages of unsupervised topic modeling with supervised string kernels for text classification tasks. The co-occurring topic words per topic and topic proportions per document obtained are used to reduce the document corpus to a topic-word sequence. This reduced representation is then used for text classification with the aid of string kernels, significantly improving accuracy and reducing training time. Experiments on the bag-of-words kernel-based string embeddings using the proposed algorithm outperform the traditional text classification approaches. This work extensively compares string kernels with topic modeling on various performance metrics to establish our findings.

The advancements in information and communication technologies contributed greatly to the development of social media and other platforms where people express their opinions and experiences. There are several platforms such as drugs.com where people rate pharmaceutical drugs and also give comments and reviews on the drugs they use and their side effects. It is important to analyze such reviews to find out the sentiment, opinions, drug efficacy, and most importantly, adverse drug reactions. Health mention classification deals with classifying such user-generated text into different classes of health mentions such as obesity, anxiety, and more. This work uses machine learning approaches for classifying health mentions from the publicly available health-mention dataset. Both the shallow machine learning algorithms and deep learning approaches with pre-trained embeddings have been implemented and the performances were compared with respect to the precision, recall, and f1-score. The experimental results show that machine learning approaches will be a good choice for automatically classifying health mentions from the large amount of user-generated drug reviews that may help different stakeholders of the healthcare industry to better understand the market and consumers.

In recent years, Native Language Identification (NLI) has shown significant interest in computational linguistics. NLI uses an author’s speech or writing in a second language to figure out their native language. This may find applications in forensic linguistics, language teaching, second language acquisition, authorship attribution, identification of spam emails or phishing websites, etc. Conventional pairwise string comparison techniques are computationally expensive and time-consuming. This paper presents fast NLI techniques based on string kernels such as spectrum, presence bits, and intersection string kernels incorporating different learners such as a Support Vector Machine (SVM), Random Forest (RF), and Extreme Gradient Boosting-XGBoost (XGB). Feature sets for the proposed techniques are generated using different combinations of features such as n-word grams and noun phrases. Experimental analyses are carried out using 8235 English as a second language articles from 10 different linguistic backgrounds from a typical NLP benchmark dataset. The experimental results show that the proposed NLI technique incorporating a spectrum string kernel with an RF classifier outperformed existing character n-gram string kernels incorporating SVM, RF, and XGB classifiers. Also, comparable results were observed among different combinations of string kernels. Interestingly, the random forest classifier outperformed SVM and XGB classifiers with different feature sets. All the proposed NLI techniques demonstrated promising results with significant improvement in training time, with the best result attaining more than a 95 percent decrease in training time. The reduced training time of proposed techniques makes it well suited to scale NLI applications for production.

Analyzing how well a movie may perform at the box office has been a hot topic in the entertainment sector for many years. Conducting research and analyzing trends is a competent skill that could lead to many commercial opportunities in the film industry. Recent technological advancements and increased data availability have made data analytics more feasible. This paper analyzes two potential commercial opportunities related to prequel movies and native language movies during the release of their respective sequel and remake movies. This work systematically analyzes data collected from the public domain on several Indian duology and native language-remake movies to unearth latent trends. Our analysis showed a distinct popularity rise for prequel and native language movies around the release of their respective sequel and remake movies. It further strengthens that if strategically planned, the rise in popularity of prequel and native language films can be formulated into business opportunities.

Depression is considered to be one of the most severe mental health issues globally; in many cases, depression may lead to suicide. According to a recent report by the World Health Organization (WHO), depression is a common illness worldwide and approximately 280 million people in the world are depressed. Timely identification of depression would be helpful to avoid suicides and save the life of an individual. Due to the widespread adoption of social network applications, people often express their mental state and concerns on such platforms. The COVID-19 pandemic has been a catalyst to this situation where the mobility and physical social connections of individuals have been limited. This caused more and more people to express their mental health concerns with such platforms. This work attempts to detect signs of depression from unstructured social media posts using machine learning techniques. Advanced deep learning approaches such as transformers are used for classifying social media posts that will help in the early detection of any signs of depression in individuals. The experimental results show that machine learning approaches may be efficiently used for detecting depression from user-generated unstructured social media posts.

In machine learning and data mining, String Kernels combined with classifiers like Support Vector Machines (SVM) show state-of-the-art results for tasks such as text classification. Traditional pairwise comparisons of strings on large datasets are computationally expensive and result in quadratic runtimes. This work compares the performance of various String Kernels and similarity measures on the document classification task. We compare different String Kernels such as Spectrum Kernel, String Subsequence Kernel, Weighted Degree Kernel, and Distance Substitution Kernel in this paper for classifying text documents. A detailed comparative study of these Kernel techniques on real-life document corpus such as Reuters-21578 shows different insights when used with and without other feature extraction techniques. The results indicate that string similarity measures give the best performance when run over the entire corpus but for small and medium-sized datasets. The complexity increases with an increase in the size of the dataset.

Blockchain, the technology infrastructure behind the famous cryptocurrency bitcoin, can take away the notion of trust from centralized organizations to a decentralized platform that is mathematically verifiable and cryptographically secure. It is gaining more significant momentum exponentially and disrupts the way businesses function beyond the digital currency aspects. This work presents a text mining literature analysis of research articles published in major digital libraries on blockchain technology and cybersecurity. This literature analysis employs automated text mining approaches such as topic modeling and keyphrase extraction for unearthing the themes from a vast body of literature. This analysis highlights the multidisciplinary nature of blockchain technology within the cybersecurity domain. The findings also show the cyber threats and vulnerabilities that evolve with blockchain technology developments. This analysis also showcases the computer security research community’s vulnerabilities and provides future research dimensions that are crucial for designing secure blockchain applications and platforms.

Social media platforms have incorporated more than half of the world's population, making it one of the most data-rich domains recently. The sentiments expressed by social media users hold great significance for various reasons, such as the identification of public opinion on a product or towards a governmental policy, to name a few. There are different domains where companies use social media sentiments to gather feedback from customers to provide them with better products and services. Only a few attempts have been reported on aspect-based sentiment analysis literature on sentiment analysis and opinion mining. This chapter proposes a framework for aspect-based sentiment analysis for social media using a topic modeling-powered approach. The experiments conducted on real-world datasets show that the proposed framework outperforms some existing works on aspect-oriented sentiment analysis.

Peer-to-peer lending is an emerging financial domain enabling people to receive instant credit facilities without much complicated procedures and intermediaries. Many financial institutions are focusing on setting up peer-to-peer lending platforms to enable hassle-free credit facilities with transparency between lenders and borrowers. The trust and transparency for hassle-free settlement and addressing novelties is a major concern in this domain which is hindering the mainstream adoption of such lending platforms. The objective of this paper is to propose a trusted and transparent distributed ledger approach using blockchain technology for setting up peer-to-peer lending platforms. The proposed approach creates cryptographically secured transactions stored over a publicly verifiable immutable ledger, which ensures credibility and auditability to investors and borrowers on every aspect of security. Proof of Immune Algorithm was proposed by leveraging the potential dendritic cell algorithm mimicking human immune system to provide consensus among peers involved in the lending process to enable trust.

In the original publication of the article, the Acknowledgements section was missed unfortunately. The correct Acknowledgements section is given in this correction.

Tracing the vaccines from the manufacturing labs till the health centers administering doses to child is a cumbersome process with the involvement of large number of stakeholders and with multiple levels of book-keeping mechanisms. This work primarily focuses on the problem of traceability of immunization vaccines from manufacturing facility, through the distribution mechanisms in cold storage network and the vaccine handling/administration facilities, till the end consumer in the Indian context. The proposed research named, Immunochain is a state-of-the-art big data and blockchain technologies powered, mobile/web-enabled vaccine traceability solution for immunization programs in India. It is vendor neutral, scalable, replicable, and reusable across different immunization programs in any geographical context. Experiments and pilot testing on Indian geography indicates that the proposed approach outperforms some of the existing approaches in vaccine traceability and record management systems.

Kernel based Deep Learning using multi-layer kernel machines(MKMs) was proposed by Y.Cho and L.K. Saul in \cite{saul}. In MKMs they used only one kernel(arc-cosine kernel) at a layer for the kernel PCA-based feature extraction. We propose to use multiple kernels in each layer by taking a convex combination of many kernels following an unsupervised learning strategy. Empirical study is conducted on \textit{mnist-back-rand}, \textit{mnist-back-image} and \textit{mnist-rot-back-image} datasets generated by adding random noise in the image background of MNIST dataset. Experimental results indicate that using MKL in MKMs earns a better representation of the raw data and improves the classifier performance.