Current scenario of the digital world is loaded with sample amounts of data and to smartly analyze this data requires researching an interesting machine learning model to suit a dataset which is a topic of research and is more focused upon. Various state of art machine learning models like supervised, unsupervised and reinforcement models along with deep learning models are available to analyze varied datasets and for a given machine learning task, selecting one model from a large pool of options is a surprisingly challenging problem, particularly if there is a lack of evaluation data to distinguish between models and this requires for a thorough understanding of the dataset which you are using to solve the task. Dataset plays an important role in the performance of any machine learning model and not always you will get a clean / well balanced dataset to directly apply the model on. The key contribution of this paper is focused on the challenge of picking a good model which can be reformulated by doing a proper analysis and cleaning of the dataset, with the objective of recommending a suitable model for a given task based on the performance of a small set of probe models under the domain of classification and regression.