Intelligent Selection of Machine Learning Algorithms - Water Tank Monitoring Example

Nowadays, Machine Learning is used in different fields as it allows to extract knowledge and predict the future based on historical data. Particularly, in the field of water tank monitoring, we noticed that the selection of the Machine Learning algorithm is arbitrary or subjectively relies on someone’s knowledge. However, the selection of the suitable Machine Learning algorithm is crucial as the inferred results may affect decision making. In the literature, some research works proposed to provide guidance for the selection of the appropriate Machine Learning algorithm. However, some of these works only suggest general guidelines and best practices. Some others, referring to meta-learning, propose to use a Machine Learning algorithm (selector) to predict the appropriate Machine Learning algorithm for a given dataset. However, in the latter works, the same addressed question arises at a higher level. Indeed, these works do not justify the selection of the suitable algorithm selector. Therefore, in this paper, we propose an approach that uses datasets from old experiments to extract knowledge about both the selector and the suitable Machine Learning algorithm for a new dataset. Moreover, the proposed approach is illustrated using an example of water tank monitoring datasets.