Journal of Artificial Neural Networks and Learning System (p-ISSN: 3049-0758, e-ISSN: 3048-6629)

Liver Disease Prediction using Multi-Layer Perceptron (MLP) Deep Learning Technique

Pooja Tiwari — 2026-06-06

As liver diseases are the causes of many health issues in the world, there is a need for accurate and efficient diagnostic methods to detect liver diseases as early as possible. In this study, a model for the classification of liver diseases using Multi-Layer Perceptron (MLP) Neural Network is developed, which is good at capturing complex and non-linear relationships in data sets. The training and testing used data comprising information on bilirubin, enzymes, and patient characteristics for liver disease. Normalization, feature selection, and discarding outliers are used to increase the accuracy of the model and avoid it becoming too fitting for the training data. It has several hidden layers with ReLU activation, and Adam is used to optimize its parameters. Common ways to assess the model are by looking at its accuracy, precision, recall, F1-score, and ROC-AUC. The experimental results reveal that the proposed MLP model has an overall accuracy of 92.26% while the accuracy of conventional machine learning algorithms such as decision tree, support vector machine and KNN are 78.57%, 83.47% and 76.32%, respectively. Moreover, the model exhibits good generalization performance on various subsets of data and is therefore relevant for clinical use. Doctors can benefit from the use of MLPs to make faster and more accurate diagnosis of liver diseases, thereby improving patients' health outcomes. Future work aims at connecting the model to real-time clinical decision aids for clinical management and further developing the framework to enable the detection of specific liver disease modalities as Hepatitis, Cirrhosis, fatty liver disease etc.

Multi-Class Lung Disease Detection Using Attention-Based Deep Learning on Chest X-Ray Images

Katam Balaji — 2026-06-30

The increasing prevalence of respiratory diseases such as COVID-19, pneumonia, and tuberculosis has significantly impacted global healthcare systems, necessitating the development of efficient, automated diagnostic solutions. Chest X-ray (CXR) imaging is one of the most widely used diagnostic tools due to its affordability, accessibility, and rapid acquisition. However, accurate multi-class classification of lung diseases remains a challenging problem because of overlapping radiographic features and variations in image quality. This paper proposes a comprehensive deep learning-based framework for multi-class lung disease detection using attention mechanisms and hierarchical classification. The proposed system integrates lung region segmentation using a U-Net architecture to isolate relevant anatomical structures and remove background noise. An attention-enhanced convolutional neural network (CNN) is employed to extract discriminative features, focusing on disease-specific regions within the lungs. Furthermore, a hierarchical classification strategy is adopted to first distinguish between normal and abnormal cases, followed by fine-grained classification of specific lung diseases. To improve model interpretability, Grad- CAM visualization is incorporated to highlight the regions influencing the model’s predictions. Experimental results demonstrate that the proposed system significantly improves classification accuracy, reduces misclassification among similar diseases, and enhances interpretability. The framework offers a reliable and efficient computer-aided diagnosis system that can support radiologists in clinical decision-making.

Deep Learning and AI Approaches for Autonomous Mobile Robot Navigation: A Simulation-Based Study with Real-World Deployment Perspectives

Ahamad Shariful Alam — 2026-06-30

Autonomous mobile robot navigation is a core research area in artificial intelligence and robotics, enabling robots to operate effectively in complex and dynamic real-world environments. Traditional navigation approaches based on geometric modeling and rule-based planning often fail to generalize in unstructured and uncertain scenarios. In contrast, recent advances in deep learning and reinforcement learning have significantly improved perception, decision-making, and control capabilities in autonomous systems. This study proposes a hybrid AI-based navigation framework that integrates convolutional neural networks (CNNs) for perception, Simultaneous Localization and Mapping (SLAM) for state estimation, and Proximal Policy Optimization (PPO)-based deep reinforcement learning for motion planning and control. The system also incorporates sensor fusion to enhance robustness under noisy and dynamic conditions. Experimental results obtained in a ROS-Gazebo simulation environment demonstrate a navigation success rate of 94%, with substantial reductions in collision rate and execution time compared to classical methods. The findings confirm that hybrid AI architectures significantly enhance adaptability, robustness, and real-time performance in autonomous navigation tasks, although challenges such as sim-to-real transfer, safety assurance, and data efficiency remain open research problems.

A Quantum Neural Network-Based Learning System for Imbalanced Data Classification

Sri Hansika Vanum — 2026-07-14

One of the biggest problems is imbalanced data classification in machine learning. The minority class is usually represented less often, causing biased predictions and a lack of generalizability. The traditional machine learning models and deep learning models maximize the overall accuracy, which results in the accuracy paradox, and the minority class detection is weak in applications like fraud detection, medical diagnosis, or anomaly detection. In order to overcome the above drawbacks, the hybrid Quantum Neural Network (QNN) learning system for imbalanced data is suggested in this paper. The proposed framework is a hybrid between classical data preprocessing and a variational quantum circuit architecture, in which the input features are encoded into qubits via angle encoding, and then processed within a layer of parameterized quantum circuits, and the output features are mapped by measurement. To enhance minority class learning in optimization, a weighted cross-entropy loss function is used. The model is tested on an imbalanced dataset and compared with several basic models, including Artificial Neural Networks (ANN), Support Vector Machines (SVM), Random Forest, XGBoost, and LightGBM. The experimental results show the improvements in minority class recall and F1-scores, achieving an overall accuracy of 94.3%, and a 7%-10% higher F1-score than baseline approaches. The results suggest that QNN-based architectures may be a suitable solution to improve the classification performance in imbalanced learning environments.

A Long-Lasting Hybrid Deep Learning Framework for Early Disease Prediction using Convolutional Neural Networks and Random Forest: Look Over and Model

Suresh Renge — 2026-07-17

As the volume of global healthcare data expands exponentially, the necessity for intelligent, sustainable predictive systems has become critical. Traditional machine learning methodologies often struggle to process the high-dimensional and heterogeneous nature of modern medical datasets effectively. To bridge this gap, this research proposes a robust Hybrid Deep Learning-Based Disease Prediction Model. Their approach synergizes the automatic feature extraction capabilities of Convolutional Neural Networks (CNN) with the ensemble classification strength of a Random Forest (RF) classifier. By substituting the standard fully connected layers of a CNN with a Random Forest, they aim to enhance generalization and reduce the risk of overfitting. When benchmarked against conventional architectures such as CNN-MLP, CNN-LSTM, KNN, and SVM, the proposed hybrid framework demonstrates superior performance. It not only achieves higher diagnostic accuracy but also contributes to the sustainability of healthcare systems by facilitating early detection and reducing the computational burden on clinical decision-making processes.