TEJAS Journal of Technologies and Humanitarian Science

Paper 1: Reducing FastText's Limits in Romanized Language Detection

Authors : Yashi Bajpai, Aditi Joshi and Mr. Amit Srivastava

Doi: https://doi.org/10.63920/tjths.44001

Abstract

To identify the language of a given text, language identification models such as FastText are used often. However, these models frequently have trouble accurately categorizing text that is written in the Roman (Latin) nature but have historically used non-Latin scripts like Hindi, Japanese and Chinese. In our research, we analyze FastText's performance on romanized inputs and find a pattern of misinterpretation into unrelated languages and lower confidence scores. We solve this by implementing a score-based thresholding method, which hides the input's anticipated language label and classifies it as romanized if the confidence score that FastText returns is less than the set threshold (0.5). This threshold-based method increases classification reliability through testing on several languages and romanized inputs. This study identifies a significant weakness in existing language identification systems and suggests a simple, adjustable modification to improve their effectiveness in multilingual, real-world situations.

Paper 2:RetentionPro-AI Powered Customer Retention & Churn Prediction System

Authors : Ravindra Chaurasia, Vaishnavi Srivastava, Sumit Chaurasiya, Shubham Singh, Anurag Singh

Doi: https://doi.org/10.63920/tjths.44002

Abstract

Customer churn—the loss of existing clients—poses a major challenge for business growth. This study predicts churn using structured datasets containing demographics, transaction history, and engagement metrics. Multiple machine learning models, including XGBoost, Random Forest, and LightGBM, were trained and evaluated using accuracy, precision, recall, and F1-score. XGBoost achieved the highest predictive performance, effectively identifying at-risk customers while minimizing false positives. The research provides a practical framework for integrating machine learning into customer relationship management systems, enabling timely interventions and data-driven strategies to improve retention, reduce churn, and enhance long-term revenue stability

Paper 3:AI Driven Health Diagnostic & Disease Prediction System

Authors : Saumya Rai, Dr. R K Singh

Doi: https://doi.org/10.63920/tjths.44003

Abstract

Proposed paper on an AI-Driven Health Diagnostic and Disease Prediction System that analyzes patient symptoms, medical history, and clinical metrics. Multiple machine learning models, including XGBoost, Random Forest, and LightGBM, were trained and evaluated using accuracy, precision, recall, and F1-score. XGBoost delivered the best predictive performance, accurately identifying high-risk patients while reducing false diagnoses. The system provides a scalable framework for integrating machine learning into healthcare platforms, enabling early detection, faster diagnosis, and data-driven clinical decision support. This approach improves patient outcomes, reduces diagnostic delays, and strengthens overall healthcare efficiency.

Paper 4:Real Time AI-Driven Threat Detection with the Integration of Zero Trust Security Framework

Authors : Shreyanshi Srivastava, Anshika Prajapati, Mr. Mahesh kumar Tiwari ,Mr. Amit kumar Srivastava

Doi: https://doi.org/10.63920/tjths.44004

Abstract

This paper elucidate how the Artificial Intelligence with Zero Trust Security determine alert dramatically threat detection and response, this offering a robust security solution against cyber threats. The cooperation of continuous verification, context based authentication, and real-time threat analysis not only reduce risks but also enhances overall security posture. This introduce the use of DL and ML techniques such as ANN and CNN in threat detection that how they mitigate with real time threats. Cybersecurity with zero trust framework and artificial intelligence is an innovative technology in real time threat detection process and adaptive response system. The amalgamation of modern Artificial Intelligence features into zero trust system is forecast to create new opportunities for enhancing adaptive security. This includes automating processes like segmentation, detecting subtle deflections or threats and refining techniques to mitigate risks. Most importantly, it strengthens the principles of Zero Trust and addresses issues associated with minimal and rule based security solutions

Paper 5:Advances and Challenges in Preprocessing Hindi–English Code-Mixed Text for Multilingual NLP

Authors : Shruti Gupta , Lakshya Srivastava , Amit Srivastava and Gaurvi Shukla

Doi: https://doi.org/10.63920/tjths.44005

Abstract

In social media and on-line communication, Hinglish is a code-mixed language between Hindi and English that is widely used in linguistically mixed areas like India. It is informally structured, it transliterates and regularly switches between languages, which poses considerable problems to natural language processing (NLP) systems. Hinglish may not be processed with the traditional preprocessing pipelines that are intended to process monolingual text. The current review offers an in-depth description of Hinglish text preprocessing and linguistic features of this language. It also talks about big datasets, benchmarks and most frequently used preprocessing algorithms like language identification, transliteration, token normalization and multilingual embeddings. The recent developments, such as contextual and code-mixed pretrained models are also mentioned. In spite of this, there are still concerns over data sparsity, annotation inconsistency, transliteration variability, and real-time processing. The paper also discusses the new areas of research, such as adaptive preprocessing systems and multiscript corpora. On the whole, this survey provides useful information on the existing developments and perspectives of strong and culturally sensitive multilingual NLP applications.

Paper 6:RideSight ML: Mobility Analytics and Forecasting

Authors : Shahnawaz , Anshuman Singh, Ankita Gupta, Ranjeet Kumar Dubey

Doi: https://doi.org/10.63920/tjths.44006

Abstract

RideSight ML is an integrated machine-learning platform designed to analyze, predict, and optimize mobility patterns across urban and regional transportation networks. Leveraging multimodal data streams—including GPS traces, public transit feeds, traffic sensors, micromobility data, and contextual variables such as weather and events—RideSight ML provides high-resolution insights into traveler demand, network performance, and system bottlenecks. The platform employs advanced statistical learning, spatiotemporal forecasting models, and graph-based neural networks to capture dynamic movement behaviors and infer latent mobility structures. RideSight ML: Mobility Analytics & Forecasting is an advanced machine-learning framework designed to transform raw transportation data into actionable mobility intelligence. Leveraging multimodal data sources—including GPS trajectories, transit schedules, shared-mobility feeds, traffic sensors, and contextual signals such as weather and events—the system employs deep learning architectures and probabilistic modeling to uncover patterns in urban movement. Core components include real-time demand prediction, dynamic travel-time estimation, anomaly detection for network disruptions, and passenger flow forecasting across modes.

Paper 7: Hybrid CNN-BiLSTM Model for Enhancing Sentiment Analysis using Text Classification on WhatsApp Group

Authors : Megha Agarwal, Vinodini Katiyar, Vandana Patel, and Bineet Kumar Gupta

Doi: https://doi.org/10.63920/tjths.44007

Abstract

The rapid expansion of social media has led to the generation of massive volumes of data, emphasizing the need to extract valuable insights, categorize information, and predict user sentiments effectively. Text classification, a prominent domain within natural language processing (NLP), focuses on organizing unstructured textual data into sentiment categories to enhance its interpretability. Achieving high accuracy in sentiment categorization calls for refined and efficient text classification techniques. Although Deep Learning models have considerably advanced this field, there remains room for optimization. This study applies the NLP framework to a WhatsApp group dataset to identify sentiment patterns and evaluates five Deep Learning models: Neural Network, Recurrent Neural Network (RNN), Long Short-Term Memory (LSTM), Bidirectional LSTM, and Convolutional Neural Network (CNN). Furthermore, it introduces a hybrid CNN–BiLSTM model that integrates feature extraction mechanisms with specific activations, dropouts, filters, kernel sizes, and layered structures to enhance sentiment prediction. The performance of the proposed architecture is benchmarked against existing research. Among individual models, LSTM and BiLSTM achieved the highest accuracy of 81 percent, while the proposed hybrid model attained an improved accuracy of 88 percent on the same dataset, demonstrating superior effectiveness in sentiment classification.

Paper 8: Advanced Cybersecurity and Surveillance Frameworks for 5G–IoT Ecosystems: Integrating Web 3.0, Blockchain, and Zero Trust Architecture

Authors : Ayushi Srivastava, Megha Agarwal

Doi: https://doi.org/10.63920/tjths.44008

Abstract

The rapid expansion of fifth-generation (5G) networks and the Internet of Things (IoT) has permanently transformed global communication infrastructures by enabling massive connectivity, ultra-low latency, and real-time data exchange. While these advancements support smart cities, healthcare, industrial automation, and intelligent transportation systems, they simultaneously introduce complex cybersecurity and surveillance challenges. Existing security frameworks often lack the scalability, intelligence, and adaptability required to protect sensitive data and ensure system resilience within dynamic 5G–IoT ecosystems. Critical issues such as real-time threat detection, device-level security, and data integrity remain inadequately addressed. The findings highlight the importance of proactive security strategies and intelligent automation in developing secure next-generation 5G–IoT ecosystems. By bridging theoretical cybersecurity models with real-world applications, this research contributes to advancing resilient digital infrastructures and promoting a secure and trustworthy digital future.

Paper 9: A Hybrid Faster R-CNN and YOLOv5 Model with Transformer Augmentation for Enhanced Object Detection

Authors : Kunal Sahu, Khushi Rajput, Shweta Sinha and Rinku Raheja

Doi: https://doi.org/10.63920/tjths.44009

Abstract

Our proposal includes a three-step model to identify small-scale objects less than 32x32 pixels, e.g., backpacks, handbags, or other discarded items in a security camera image. We initially determine potential boxes with YOLOv5. Then we fine-tune those boxes with Faster R-CNN to achieve more precise results. We now include a small Transformer decoder to detect smaller objects. We will prune the model using weight pruning and INT8 quantization, and will make the size of the model smaller by 20-30%, and targeting 20-30 frames per second on a Jetson Nano to make it executable in real time. Our training will be done on mixed precise on a custom surveillance set which concentrates on small things. We are aiming to make the recall of small objects exceed the 30% baseline by YOLOv5 with obvious benefits in autonomous car, smart security, and farm monitoring applications. The model will subsequently be tested on our set by running the model later, testing it on COCO and KITTI, and testing its ability to work with video streams.