A Higher Accuracy Classifier Based on Semi-supervised Learning

Abstract

Mining data has attracted many researchers because of its usefulness of extracting valuable information from the huge volume of continuously increasing databases. In general using labeled data has been more difficult and time consuming than using unlabeled samples. There are several methods that could be used to build a classifier using unlabeled samples. However these may suffer from poor classification quality. In this paper, we propose a semi-supervised approach of classification which uses fewer amount of labeled data and large amount of unlabeled data to build a higher accuracy classifier. Various standard and synthetic dataset have been used to demonstrate the effectiveness of our approach.

Using supervised learning to classify authentic and fake online reviews

“…Data were analyzed using supervised learning, which includes machine learning algorithms that use labeled data for training and testing [29]. Related studies have used various supervised learning algorithms.…”

“…Supervised learning refers to machine learning techniques in which models are trained with data containing class labels. After the model is supervised to learn, it is tested on unknown data for performance evaluation [29].…”

Online Data Stream Learning and Classification with Limited Labels

“…The proposed algorithm uses a batch learning approach that does not include online classification and incremental learning capability. Reference [9] proposed a kmeans clustering with retraining mechanism. It performs online classification and retraining to handle concept drift.…”