Machine learning methods involve both supervised and unsupervised learning. On the other hand, semi-supervised and self-supervised learning also play a crucial role, with many applications increasingly relying on these paradigms. Real-world applications such as healthcare, marketing, and digital media generate massive data. However, labeling this data is often impractical due to resource constraints. Consequently, there is an increasing need for techniques that can efficiently utilize limited labeled data to facilitate the labeling of large amounts of unlabeled data. Semi-Supervised Clustering (SSC) addresses this challenge by using available labeled data to guide the clustering process. However, maintaining clustering quality and handling outliers remain significant challenges. To tackle these issues, we proposed Radial Expansion Compact Clusters (RECC), a novel framework based on the label-based clustering paradigm to form quality compact clusters while efficiently assigning labels to unlabeled data. RECC also incorporates an outlier detection mechanism to minimize the impact of misleading data on clustering process. Extensive experiments on real-world datasets demonstrate that the proposed RECC method outperforms the conventional methods in improving clustering quality.

Semi-supervised learning plays an important role by combining both labeled and unlabeled data, bridging the gap between supervised and unsupervised learning. In today's fast-changing data streams, an efficient classifier is essential to adapt to evolving patterns. It is also important to address challenges like concept drift and concept evolution to maintain consistent model performance. However, handling unlabeled data in semi-supervised data streams can be challenging, as it requires significant resources and can be costly. This paper introduces an Incremental Workflow Model (IWM) that effectively handles the underlying issues of the data stream in a semi-supervised environment while using minimal labeled information. The proposed model is designed to be highly adaptable, ensuring it can handle the concept drift and concept evolution challenges. One-Class Classifier (OC_Classifier) with a proposed confidence score is used to identify the novel data and to predict known data in the stream. Further, MicroClusters are employed to reduce the memory overhead in the Ensemble Model. Experiments are performed on benchmark datasets, and results are analyzed to demonstrate the effectiveness of the proposed workflow model.

Due to rapid growth in information technologies, millions of data are being generated at every time span. Data storage and processing are more expensive in terms of memory, resources, and time. The traditional machine learning classifier performs well when all the data is available along with the label information at the time of training, which may not be possible in every circumstances. Semi-supervised techniques overcome this limitation. With limited labeled data, semi-supervised learning makes use of both classification and clustering techniques for constructing efficient classifiers, and there is a demand to update the classifier periodically. Unlike traditional approaches, which use the entire data set for updating the classifier, our method updates in an incremental fashion. This paper proposes a framework for incremental classifiers in a semi-supervised environment. Experiments are conducted on benchmark datasets with the proposed approach and compared to the conventional method.