Prognostic modeling is essential for forecasting future clinical scores and enabling early detection of Alzheimer's disease (AD). While most existing methods focus on predicting the ADAS-Cog global score, they often overlook the predictive value of its$\mathbf{1 3}$sub-scores, which reflect distinct cognitive domains. Some sub-scores may exert greater influence on determining global scores. Assigning higher loss weights to these clinically meaningful sub-scores can guide the model to focus on more relevant cognitive domains, enhancing both predictive accuracy and interpretability. In this study, we propose a weighted Vision Transformer (ViT)-based multi-task learning (MTL) framework to jointly predict the ADAS-Cog global score using baseline MRI scans and its 13 sub-scores at Month 24. Our framework integrates ViT as a feature extractor and systematically investigates the impact of sub-score-specific loss weighting on model performance. Results show that our proposed weighting strategies are group-dependent: strong weighting improves performance for MCI subjects with more heterogeneous MRI patterns, while moderate weighting is more effective for CN subjects with lower variability. Our findings suggest that uniform weighting underutilizes key sub-scores and limits generalization. The proposed framework offers a flexible, interpretable approach to AD prognosis using end-to-end MRI-based learning. (Github repo provided11https://github.com/mirahhamid/A-Weighted-Vision-Transformer-Based-Multi-Task-Learning-Framework-for-Predicting-ADAS-Cog-Scores.git)

Human activity recognition is a crucial field of study, but current approaches often require ground truth labels, which are not always available. We propose a new method called the Flexible Multi-Objective Particle swarm optimization clustering method based on Game theory (FMOPG), which can identify human activities without any supervision. Unlike traditional clustering methods that require an estimate of the number of clusters and are often inaccurate, FMOPG handles varying cluster numbers with an incremental technique, selecting clusters with good connectivity and separation. We enhance Particle Swarm Optimization (PSO) with mean-shift vectors for faster convergence and better handling of non-spherical clusters. Employing multi-objective optimization and Gaussian mutation, FMOPG provides a range of optimal solutions. We map the optimization problem to game theory to select the best solution based on different criteria. A smart grid-based method is proposed for population initialization, reducing variance and improving reliability. FMOPG outperforms state-of-the-art methods, improving clustering accuracy by 3.65%. Moreover, the incremental technique has improved clustering time by 71.18%.

Despite many advances in Human Activity Recognition (HAR), most existing works are conducted with supervision. Supervised methods rely on labeled training data. However, obtaining labeled data is difficult, costly, and time-consuming. In this paper, we introduce an automatic multi-objective particle swarm optimization clustering based on Gaussian mutation and game theory (MOPGMGT) to provide fully unsupervised human activity discovery. Furthermore, we map the multi-objective clustering problem to game theory to get the best optimal solution. The proposed algorithm can accurately find the number of activities without any prior knowledge. Multi-objective optimization problems typically cannot have a single optimal solution. We solve this problem by applying, Nash Equilibrium (NE) to the pareto front as the decision-making for choosing the best solution. NE does not just look for the best solution but tries to optimize the final solution by considering the effect of choosing each of the solutions as the best solution on the other solutions and one with the best impact is chosen. Moreover, a Gaussian mutation is applied on the pareto front to avoid premature convergence. As far as we know, this is the first time that human activity discovery is performed fully unsupervised, and a multi-objective PSO is mapped to the game theory space for finding the best solution. Experiments on six challenging human activity datasets demonstrate the capability of the proposed approach in achieving the best accuracy in human activity discovery and determining the optimal number of clusters. In comparison to well-known multi-objective algorithms, the MOPGMGT significantly improves the clustering outcomes on six benchmark clustering datasets.

Human activity discovery aims to cluster human activities without any prior knowledge of what defines each activity. However, most existing methods for human activity recognition are supervised, relying on labeled inputs for training. In reality, it is challenging to label human activity data due to its large volume and the diversity of human activities. To address this issue, this article proposes an unsupervised framework for human activity discovery in 3-D skeleton sequences. The framework includes a data preprocessing step that selects important frames based on kinetic energy and extracts relevant features, such as joint displacement, statistical displacement, angles, and orientation. To reduce the dimensionality of the extracted features, the framework uses principle component analysis. Unlike many other methods for human activity discovery, the proposed framework is fully unsupervised and does not rely on presegmented videos. To segment the time series of activities, the framework uses a sliding time window with some overlapping. The hybrid particle swarm optimization (PSO) with Gaussian mutation and K-means algorithm is then proposed to discover the activities. PSO is chosen for its powerful global search capability and simple implementation. To further improve the convergence rate of PSO, K-means is applied to the outcome centroids from each iteration of PSO. The experimental results on five datasets demonstrate that the proposed framework has superior performance in discovering activities compared to other state-of-the-art methods. The framework achieves an average increase in accuracy of at least 4%.

While there has been many new developments in multiobjective evolutionary algorithms, they have not been applied or investigated in clustering problems. In this paper, ten different unsupervised clustering techniques applying different MOEA (SPEA2, IBEA, MOEA/D and MOEA/GLU), PSO (QPSO and BBPSO) and Fuzzy approaches are experimented on ten public datasets. The rationale to apply MOEA is to increase the exploitation capabilities of clustering techniques to further refine the cluster solutions found by fuzzy or PSO clustering. The aim is to investigate in the performance of different types of MOEA applications in clustering, determining whether MOEA Fuzzy clustering outperform MOEA PSO variants. Overall, MOEA/D BBPSO was found to produced the best results. It outperformed MOEA Fuzzy techniques, having tested on datasets with high number of classes, that are imbalanced and/or overlapping classes. IBEA Fuzzy clustering was found to produce the worst results. MOEA/D clustering was found to perform better than other MOEA techniques. In this work, we showed that MOEA/D BBPSO clustering produced the best results on challenging datasets. It was able to use MOEA/D to deepen its exploitation capability while benefiting from the exploratory ability of BBPSO when clustering challenging datasets.

The idea behind distance metric learning (DML) is to accentuate the distance relations found in the training data, maintaining whether the data patterns are similar or dissimilar. In this paper, we investigate in using DML (GDML, LMNN, MCML and NCA) in semi-supervised Fuzzy c-means clustering and apply them on a real, biomedical dataset and on UCI datasets. We used a cross validation setting with varying amount of labelled data to test our methodology. Out of eight datasets, statistical significant improvement was found on five datasets using ssFCM with DML. This shows that DML can improve ssFCM clustering for some datasets. Further analysis using 2D PCA projection and sum of squared distances before and after DML transformation of the original data are carried out. Interestingly, DML was found to worsen ssFCM clustering in the NTBC dataset with hierarchical clusters.

The aim of this work is to identify clinically-useful and stable breast cancer subgroups using a reduced panel of biomarkers. First, we investigate the stability of subgroups generated using two different reduced panels of biomarkers on clustering of breast cancer data. The stability of the subgroups found are assessed based on comparison of agreement levels using Cohen's Kappa Index on clustering solutions from ssFCM methodologies, consensus K-means and model-based clustering. The clustering solutions obtained from the feature set which achieve the higher agreement is chosen for further biological and clinical evaluation to establish the subgroups are clinically-useful. Using a ssFCM methodology, we identified seven clinically-useful and stable breast cancer subgroups using a reduced panel by Soria et al. So far, the stability of the subgroups identified using the reduced panel of biomarkers have not yet been investigated.

In previous work, six clinically novel and useful subgroups of breast cancer were identified using rules and clinicians' expertise to combine solutions from three different clustering algorithms on a database of biomarkers. The motivation for the present work is to reproduce this classification using a single clustering method. In the long term, we hope to produce a clinically useful classification using fewer features (biomarkers), reducing the time and cost of running complex and expensive clinical tests. Hence, the aim of this paper is to investigate the use of feature selection in combination with ssFCM to reduce the number of features while maintaining accuracy (defined as agreement with the previous classification), both on our breast cancer biomarker data and on other benchmark datasets. We show experimental results using four feature selection techniques, exploring with 10, 15 and 17 selected features out of the original 25 biomarkers for breast cancer. We experimented with varying amounts of labelled data (10% - 60% of the training data) and we evaluate classification accuracy using cross-validation. It was found that classification accuracy increased using 15 or 17 breast cancer biomarkers. Using SVM-RFE and CFS, improved classification accuracy was found on three UCI datasets, Arrhythmia, Cardiotocography and Yeast.

There are many issues to be considered in the design of distance-based fuzzy semi-supervised clustering (FSSC) algorithms. To identify these issues, we compare the performance of four such algorithms. We describe the properties of these algorithms, highlighting their key differences, and then experimentally compare their performance on common datasets. Several experimental conditions are investigated. Firstly, two forms of initialisation of the membership values of unlabelled patterns are used; 1/c and 0. Secondly, the algorithms are run with varying proportions of labelled patterns in the datasets, ranging from 2% to 40%. We find that no algorithm outperforms the others in all the datasets. We also observe that small modifications in similar objective functions can improve clustering, and that most of the algorithms perform slightly better with zero initialisation of unlabelled patterns. An interesting observation is that the increase in labelled patterns does not always improve clustering. From these results, we conclude that the number and scale of dimensions in the data set, initial partition matrix, distance metrics and objective functions, together, affect clustering results. In addition, we conclude that not all initially labelled patterns are good candidates for supervision.