EAI Endorsed Transactions on Intelligent Systems and Machine Learning Applications

Detection, Localization of Cardiomegaly and TB Disease of CXR Images using Deep Learning

Ganesh Pradeep P V — 2024-07-08

INTRODUCTION: Tuberculosis (TB) continues to pose a significant worldwide public health concern, as it stands as the primary contributor to mortality stemming from infectious illnesses. Cardiomegaly, characterized by an enlarged heart, poses medical concern as well.

OBJECTIVES: Timely identification of Cardiomegaly is vital for effective management. Chest X-ray diagnosis is an easily available method with less radiation exposure to detect several lung infections and heart enlargement. Utilizing computer-aided diagnostic systems can aid in the early detection of lung conditions and the enlargement of the heart.

METHODS: We worked on different state-of-the-art CNN architectures such as VGG, DenseNet and EfficientNet with customization over dataset generated from combination of multiple publicly available datasets, which consists of 12939 annotated images across three different categories, one being normal and other two being TB and cardiomegaly diseases..

RESULTS: EfficientNetB5 with optimization has shown excellent results amongst others in classifying Tuberculosis and Cardiomegaly with a remarkable accuracy of 97%.

CONCLUSION: The proposed model is ready for clinical diagnosis and triaging of X-ray images. Our solution also offers efficient ways to show the presence of the above diseases using Grad-CAM technique.

Enhancing Document Clustering with Hybrid Recurrent Neural Networks and Autoencoders: A Robust Approach for Effective Semantic Organization of Large Textual Datasets

Ratnam Dodda — 2024-03-18

This research presents an innovative document clustering method that uses recurrent neural networks (RNNs) and autoencoders. RNNs capture sequential dependencies while autoencoders improve feature representation. The hybrid model, tested on different datasets (20-Newsgroup, Reuters, BBC Sports), outperforms traditional clustering, revealing semantic relationships and robustness to noise. Preprocessing includes denoising techniques (stemming, lemmatization, tokenization, stopword removal) to ensure a refined data set. Evaluation metrics (adjusted randomness evaluation, normalized mutual information evaluation, completeness evaluation, homogeneity evaluation, V-measure, accuracy) validate the effectiveness of the model and provide a powerful solution for organizing and understanding large text datasets.

Influence of Promotion and Pricing on Purchase Incidence, Demand, and Sales Using Machine Learning

Rahul D Shanbhogue — 2024-04-10

The consumer goods industry is a dynamic and fast-paced sector that faces significant challenges in meeting the consumer’s ever-evolving demands and preferences. Today’s retail businesses focus on optimizing their supply and retail execution to maintain a competitive edge in the market and remain profitable. The most impactful method is to offer promotional events that stimulate large-scale purchases and attract new customers. The patterns of normal sales days, promotion days, and non-promotion days are different and it is vital to capture the influence of promotions on demand and sales. Thus, it is vital to understand the effects of promotion and plan them. This paper aims to understand the influence of promotion and pricing strategies for FMCG retail businesses to maximize demand for each brand. Explore the use of Machine Learning (ML) and Deep Learning models such as Clustering and Neural Networks to identify and understand the various demand patterns to analyse the influence of promotion and pricing on demand, and enable businesses to respond more quickly to changes in the market by enabling them to make better-informed decisions that can mitigate risks associated with the impact of disruptions and to ensure the continuity of the business.

Diagnosis of Glioma, Menigioma and Pituitary brain tumor using MRI images recognition by Deep learning in Python

Seyed Masoud Ghoreishi Mokri — 2024-04-15

Medical image processing is a very difficult and new field. One thing they do in this field is analyze pictures of people's brains to look for signs of tumors. They use a special computer program to help with this. This paper talks about a new way to use the program to find brain cancer early by looking at the texture of the tumor. This paper explains how we can find and understand brain tumors using special pictures called MRI scans. We use computer programs to help us do this. First, we find the tumor, then we separate it from the rest of the brain, and finally we measure how big it is. We can also figure out how serious the tumor is by looking at different kinds of tumors. To make it easier for people to use, we made a special program in a computer language called COLAB for python codes about using CNN network for deep learning. We tested this program on 8 patients and learned a lot about their tumors.

Detecting Alzheimer’s Patients using Features in Diﬀerential Waveforms of Pupil Light Reﬂex for Chromatic Stimuli

Minoru Nakayama — 2024-07-18

A procedure to detect irregular signal responses to pupil light reflex (PLR) was developed to detect Alzheimer’s Disease (AD) using a functional data analysis (FDA) technique and classification with an Elastic Net. In considering the differences in features of PLRs between AD and normal control (NC) participants, signals of summations and differentials between experimental conditions were analysed. The coefficient vectors for B-spline basis functions were introduced, and the number of basis was controlled for an optimised model. Model trained data was created using a data extension technique in order to enhance the number of participant observations. In the results, the required number of basis functions for differential signals is larger than the number for the their summation signals, and the features of differential signals contribute to classification performance.

Exploring The Efficiency of Metaheuristics in Optimal Hyperparameter Tuning for Ensemble Models on Varied Data Modalities

Vivek BC — 2024-08-06

Effective disease detection systems play an important role in healthcare by supporting diagnosis and treatment. This study provides a comparison of hyperparameter tuning methods for disease detection systems using four health datasets; kidney disease, diabetes detection, heart disease and breast cancer detection. The main objective of this research is to prepare datasets by normalizing the input and testing machine learning models such as Naive Bayes Support Vector Machine (SVM), Logistic Regression and k Nearest Neighbor (kNN). to identify effective models for each data set. After implementing the models, we apply three hyperparameter tuning techniques: Grid search, random search, and particle ensemble optimization (PSO). These methods are used to tune the model parameters. Improve overall performance metrics. The evaluation focuses on accuracy measurements to compare model performance before and after hyperparameter tuning. The results of this study illustrate how different tuning techniques can improve the performance of disease detection systems across a range of healthcare datasets. By conducting testing and analysis, we determine the appropriate tuning method for each data set, yielding valuable insights, to develop an accurate and effective disease detection system .These discoveries serve to advance the field of healthcare analytics and machine learning to deliver outcomes for patients and healthcare services.

Deep Learning Approaches for Stock Price Prediction A Comparative Study on Nifty 50 Dataset

Sushma P Kallimath — 2025-02-28

Stock price prediction is essential for investors and traders in financial markets. Deep learning methods have emerged as promising tools for capturing intricate patterns in stock market data. In this paper, we explore a comprehensive comparative study of various deep learning architectures for stock price prediction using the Nifty 50 dataset. The models evaluated include Linear regression, LSTM, GRU, CNN, RNN, Temporal Convolutional Network (TCN), as well as combination models such as LSTM+GRU, CNN+RNN, CNN+TCN, and LSTM+TCN. Our study aims to evaluate how well they perform and suitability of these methodologies in capturing the dynamics of stock price movements. Utilizing historical Nifty 50 data spanning multiple years, we evaluate the models’ predictive capabilities using standard evaluation metrics such as MSE, R2 Score, RMSE, MAE, and MAPE. Results from our experiments unveil distinct strengths and weaknesses among the different deep learning architectures. While linear regression provides a baseline for comparison, deep learning models like LSTM, GRU, CNN, RNN, and TCN exhibit superior performance in capturing the nonlinear and time-varying nature of stock market data. Additionally, hybrid architectures demonstrate promising results by leveraging the complementary strengths of individual models. This comparative study offers meaningful perspectives on the effectiveness of various deep learning approaches for stock price prediction, which can benefit researchers, practitioners, and stakeholders in the financial domain. By understanding the performance characteristics of these models, stakeholders can make informed decisions in their investment strategies.

Modelling attention levels using microsaccade rates in response to vibrations in the peripheral ﬁeld of vision

Minoru Nakayama — 2025-03-18

Viewer’s eye movements and behavioural responses were analysed in order to determine the relationship between selective perception and visual attention during a dual detection task in the central and peripheral ﬁelds of vision, in order to design better functioning information displays. Changes in visual attention levels were evaluated using temporal microsaccade rates. Response accuracy of stimulus detection and the microsaccade rate were analysed using a hierarchical Bayesian modelling technique. In the results, the dominance of the response in the peripheral ﬁeld of vision is conﬁrmed to be deviations in the estimated parameters. Also, chronological changes in levels of attention and the contribution of these changes to behavioural responses were examined. The relationship between behavioural responses, microsaccade rate, and the directional dominance of certain viewing areas in the peripheral ﬁeld of vision were discussed, in order to evaluate the level of visual attention of viewers.

Optimising performance indicators in the telecommunications sector

Fiston Chrisnovic Balanganayi Kabutakapua — 2025-06-06

INTRODUCTION: This study analyses and predicts performance in the public telecommunication sector using neural networks on key performance indicators in the telecommunication sector.

OBJECTIVES: Although there are several key performance indicators in the telecommunications sector, we have selected a few and assessed their correlation with the variable to be predicted. In this study, we used supervised learning based on a multi-layer neural network.

METHODS: The algorithm used in this study is the retro propagation algorithm because of its simplicity and accuracy of estimation.

RESULTS: The results show that the selected indicators, including accessibility, maintainability, satisfaction, network operating cost and availability, explain more than 80% of the performance in the telecommunications sector, and the area of the ROC curve is equal to 0.97, which means that the classifier is almost perfect. This is also justified by the sensitivity and specificity, which are close to 1 when observing the ROC curve and the confusion matrix. The classification error found from the confusion matrix is equal to 1%, which means that our model has very high accuracy.

CONCLUSION: The other indicators presented were not selected in the model because of their low correlation with the variable of interest and the difficulty of collecting the data.

Aspects-based representative significance of Machine Learning algorithms & natural language processing applications in nanotechnology.

Pascal Muam Mah — 2024-10-25

Introduction: The rapid changes in computational power of machine learning algorithms and natural language processing applications have led to multi-scale and many core designs in nanotechnology. Machine learning algorithms and natural language processing applications are easing the burden engineers have to go through to understand nanoparticles.
Problem: There is still a challenge to predict and control particles of nanomaterials at nanoscale. Aspect-based climatic conditions are negatively impacting the world with huge modification on nanoparticles, nanomaterials and nanostructures.
Objective: Study examines aspects of machine learning algorithms and natural language processing applications that can be used to predict and control particles, and structure of nanomaterials at nanoscale. Method and materials. The study examines significance of machine learning algorithms & applications in nanotechnology, examines aspects of machine learning algorithms & natural language processing applications applied in nanotechnology, and discusses current-future trends of nanotechnology based on learning algorithms & natural language processing applications.
Results and conclusions. The findings result in the conclusion that machine learning & natural language processing application in nanotechnology is implementing an advanced microscopic revolution with the potential to metamorphose the world's industrialization and scale human existence. Machine learning algorithms have the potential to predict and classify nanomaterials and natural language processing has the potential to retrieve relevant data hidden within the classified nanomaterials which results has a huge significance in the pharmaceutical industry