DSSL is a machine learning paradigm that performs classification of data, similar to supervised learning. The difference is that DSSL utilizes both unlabeled and labeled data. DSSL algorithms have proven extra effective especially when the dataset contains a small portion of labeled datapoints and a large portion of unlabeled datapoints. The extensive use of additional unlabeled data will increase the computational time of each algorithm exponentially. Thanks to GPUs the computational time of the algorithms will be reduced drastically. Many of these algorithms we wish to use have previous only been evaluated on benchmark datasets. In this research the algorithms are evaluated on real-world datasets. Applying algorithms to real-world datasets and get the same accuracy as on benchmarks is difficult and several changes has to be made on the algorithm. Therefore, many test runs need to be performed.