International Research Journal of Multidisciplinary Technovation

A Systematic-Architectural-Perspective Based Performance Analysis of A-MERIT-C- Dynamic Learning Multitiered Ensemble-Based Real Time Flight Data Analysis

Shailaja B. Jadhav, Kodavade D.V, Nagraj V. Dharwadkar — Thu, 30 Oct 2025 00:00:00 +0000

Large-scale data analysis has been the subject of numerous studies recently. In many applications of today's data-intensive world, data is typically brought in continually as data streams. Analytics engines that handle streaming data must be able to react to data that is in motion. Data streams provide special challenges because traditional methods for data mining and machine learning are meant for static information. They are less suited to consider the representative characteristics of data streams and are very less suitable to effectively analyse data that is growing quickly. The authors through this research viz. A-MERIT-C - a dynamic learning multitiered ensemble-based flight real time data analysis system. Through this research authors have presented an active learning dynamic real time data stream analysis model built with self-tuning ensemble learning framework, able to quickly adapt to concepts in near real time streaming data analysis. The conceptual architectural framework illustrated through this research is adaptive to deal with the dynamics related with real time data through the evolving classifier pool (i.e. best performing classifiers get added to classifier pool at every epoch). One more distinguishing characteristic of -A-MERIT-C is instead of using traditional hold out evaluation, it uses prequentially evaluated classifiers. A-MERIT-C's unique features provide significant gains in accuracy, precision, and AUC for streaming data analytics; however, it can also overcome the drawbacks of current algorithms, including concept evolution and feature drift, by using incremental learning and feedback.

A Unified Model for Estimation of Reference Evapotranspiration Using an Assembly of Ensemble Learners Coupled with Swarm Intelligence Optimizers

Gouravmoy Banerjee, Uditendu Sarkar, Indrajit Ghosh — Tue, 14 Oct 2025 00:00:00 +0000

Several machine learning models and their ensembles have been suggested for reference evapotranspiration (ET₀) modeling at different climatic regions. Researchers reported that optimizing model hyperparameters using an intelligent algorithm significantly improves the performance of such models. However, ensemble models hybridized with hyperparameter optimizers have hardly been applied for the precise estimation of ET₀ worldwide. The current research is devoted to designing sixteen hybrid versions of four ensemble models, alternatively coupled with four popular swarm intelligence optimization algorithms and finding the best-fit model against different input combinations of available climatic parameters for the groundwater-stressed region of North Bengal, India. The performances of four ensemble models and their sixteen hybrid versions were compared in terms of four well-recognized statistical metrics: the coefficient of determination (R²), Nash-Sutcliffe efficiency (NSE), root mean squared error (RMSE), and mean absolute error (MAE). Experimental results depicted that in nearly 92% of cases, the hybrid versions outperformed the primary ensemble models, irrespective of the available climatic parameters. In most cases, the ensemble models hybridized with the whale optimization algorithm (WOA) produced the highest estimation accuracy, followed by the sailfish optimizer (SFO). Solar radiation was also found to be the most significant climatic parameter for estimating ET₀ in this region.

YOLO-Based Generalized Framework for Leukemia Cell Detection Using Unified Microscopic Image Datasets

Ratnamala Mantri (Paswan), Rais Abdul Hamid Khan — Wed, 15 Oct 2025 00:00:00 +0000

Acute lymphoblastic leukemia is a kind of blood cancer that attacks the lymphoblast, a subgroup of white blood cells. Leukemia is a potentially lethal hematological cancer that requires prompt diagnosis. A skilled manual blood smear examination is one of the laborious and prone to human error conventional diagnostic methods. Although current automated methods developed by researchers use either single-cell or multi-cell pictures to detect leukemia cells, they frequently lack model generalization that perform better on heterogeneous datasets. They are also insufficient for deployment in real time. This study aims to develop generalized real-time system for detecting ALL cells from single and multi-cell microscopic blood smear images. The system utilizes three YOLO based state-of-the-art models: YOLO11, YOLOv8 and YOLOv5. The core novelty of this study lies in the creation of a unified dataset that integrates both single-cell and multi-cell microscopic blood smear images, this enables the model to learn generalized representations from diverse image contexts. Three datasets are merged to create the unified dataset, ALL-IDB1: multi- cell images, ALL-IDB2 & C-NMC-19: single-cell images. Image annotation and preprocessing are performed using Roboflow platform, while Google Colab is used for training and testing. These models are trained separately on individual datasets and the unified dataset.The performance of generalized YOLO models is assessed and contrasted against dataset-specific models using mAP@50 and recall metrics on the same set of unseen images from all three datasets.The experimental results indicate that generalized YOLOv8 model achieved notably high recall and competitive map@50, demonstrating strong adaptability and accuracy. These results highlight YOLOv8 as a promising solution for developing generalized model for leukemia cell detection.

Detection and Classification of Genetic Acute Myeloid Leukemia Cells using Deep Learning Techniques

Hema Patel, Himal Shah, Gayatri Patel, Atul Patel — Thu, 16 Oct 2025 00:00:00 +0000

Acute Myeloid Leukemia (AML) is a hematological disease that is defined by the fast growth of aberrant myeloid precursor cells in the blood and bone marrow, which disrupts normal hematopoiesis. Treatment and prognosis are influenced by the early detection of this deadly illness and its appropriate classification. Therefore, utilizing the Human Leukemia Cytomorphology Collection dataset, which comprises leukemic and normal single-cell images of Acute Myeloid Leukemia (AML) type, this research suggests a deep learning-based hybrid model for automated leukemia detection and classification. By taking morphological characteristics and genetic abnormalities into account, leukemic cells have been distinguished. The features in this study are extracted using MobileNetV2, ResNet-101, and VGG-16. Then, feature-level stacking is performed using the Support Vector Machine and Random Forest classifiers for final classification using Principal Component Analysis (PCA). Utilising image segmentation, normalisation, data augmentation, and data oversampling techniques, the pipeline improves data quality and corrects class imbalance. Additionally, t-distributed Stochastic Neighbour Embedding (t-SNE), which shows the extracted features used for the detection of leukemia subtypes, and Gradient-Weighted Class Activation Mapping (Grad-CAM) images help with interpretability by highlighting important decision areas. The suggested study achieved 98.35% accuracy, 95.87% precision, 95.84% sensitivity, 98.97% specificity, and 95.74% F1-Score. Along with the trial results, a comparison of the four separate frameworks, viz., MobileNetV2, ResNet-101, VGG-16, and Vision Transformer, has also been carried out. The comparison shows that the proposed model outperforms the other frameworks. The outcomes show that the suggested model has the capability to be used a reliable means for the prompt identification of AML and its subtypes.

Evaluating and Improving Algorithms for Off-Road Routing

Sat, 25 Oct 2025 00:00:00 +0000

In this paper, we explore the multifaceted capabilities of route-finding algorithms and their role in delivering dynamic paths for diverse navigation scenarios. In off-road route finding, the shortest path is not always the best; the route's smoothness must also be considered. Present methodologies like A* have notable limitations, such as difficulty adapting to complex terrains and producing rugged routes. This limitation hampers performance in critical scenarios, such as in emergencies like landslides or earthquakes, where off-road exploration is crucial. This paper proposes 3 methodologies that have been fine-tuned with optimal hyperparameters for optimal results using gradient descent. Two of these methodologies, Simulated Annealing and Ant Colony Optimization, overcome some limitations of A* but not all. However, Q-Learning significantly overcomes the limitations and saves travel time by providing a route with 70% fewer undulations and a 16% reduction in route length compared to A*. Compared to other implementations, the Q-Learning implementation proposed in this research not only focuses on minimizing path length but also minimizes route undulations, providing a dual objective approach that is well suited to real-world scenarios. Unlike prior implementations, which focus on a single objective, such as path length or obstacle avoidance, this work leverages a reward function that penalizes elevation variance while rewarding shorter routes, resulting in smoother, more easily traversable paths. Thus, Q-Learning overcomes the cons of the present methodologies and can form synergistic combinations, enhancing the overall performance of off-road space searching systems and accommodating the various challenges and complexities inherent in the targeted applications. The dataset used includes features such as latitude, longitude, and elevation. The strategic application of heuristics enables the swift evaluation of multiple paths, facilitating the selection of optimal routes in real-time applications. By combining various heuristics, the development of off-road path identification systems capable of discerning optimal paths across varied terrains becomes feasible.

A Multi-Label Toxic Comment Classification Framework using under Sampling and Deep Learning Models

Sushma S, Vamsi Krishna M, Sasmita Kumari Nayak — Mon, 27 Oct 2025 00:00:00 +0000

Toxic online content poses significant challenges to digital communication platforms, necessitating accurate and balanced classification strategies. Unlike traditional binary classification approaches, this study focuses on multi-label toxic comment classification using the Jigsaw dataset, where each comment may exhibit multiple overlapping toxicity types. To address the severe class imbalance inherent in the dataset, three tailored undersampling strategies—One-vs-Rest Undersampling, Multilabel Random Undersampling (MLRU), and an Improved Threshold-Based Undersampling—are proposed to enhance the representation of minority labels such as threat and identity hate. These undersampled datasets are evaluated using a diverse set of ML and DL models, including Random Forest, XGBoost, CNN, RNN, LSTM, BiLSTM, BERT, and RoBERTa. Experimental results shown that this joint multi-label–specific under sampling combined with advanced classification architectures establish superior models, more evident in early detection of rare but relevant types of toxicity. This work demonstrates that multi-label learning frameworks can serve as an effective approach toward fair and full toxic comment detection.