An autoML tool for software vulnerability prediction

The popularization of machine learning has been inevitable, especially in recent years, as large amounts of data have become more accessible and provide significant benefits for industries such as finance, healthcare, logistics, and technology. The need to implement machine learning applications, which have started to be used in different sectors, has increased rapidly and efficiently. However, the deep technical knowledge and coding skills required to build a machine-learning model using traditional methods may not be accessible to everyone. Therefore, no-code or low-code platforms have started to attract more attention. These platforms allow users to load datasets, define features for building the model, and even train and evaluate the model with non-technical interfaces. In this way, it eliminates the need to write complex code, making it possible for a wider audience to be involved in machine learning projects. In this thesis, a platform that enables performing no-code machine learning prediction models for software vulnerability analysis is developed, which offers important features such as run-time limiting and model degradation analysis in addition to other no-code platforms. This platform is named PPDaaS (Prediction Platform as a Service) and is used to train prediction models for "Software Vulnerability" and "OpenML" benchmarking datasets. Initially, the platform is tested with benchmarking data that is used for AutoML platforms. The performance of the classification algorithms is evaluated by placing a time limit on the "OpenML adult" dataset, to make a fair comparison between the machine learning algorithms. Once the platform is tested for its intended AutoML functionalities, it is used to find model degradation on Software Vulnerability data. Finally, the feedback about the developed PPDaaS platform is gathered from end users as well. The results indicate that the developed PPDaaS platform can find a place for itself in the software industry.