Data analysts perform exploratory programming for several analytical tasks on notebooks. One is Data Discovery which consists in finding attributes that might join. This is timeconsuming and new techniques are needed to provide joinable attributes and receive a speed-up to analyse data. Those attributes should produce high quality joins. We consider high quality joins those joins between attributes that share a high number of unique values. In this thesis, we aim to find quality joinable attributes by proposing a three-step approach: performing attribute profiling, classification and ranking. We create 5 categorical labels to represent the quality join that two attributes might have. One-vs-the-Rest strategy is used to create machine learning models. We aim at integrating data discovery with notebooks and well-known data management tools. We prototype our techniques on top of mature tools for exploratory and large-scale data processing, namely Jupyter and Apache Spark. We created four experiments with real datasets to validate our approach. Our experiments suggest our approach is a general approach for finding high quality joins for any topic. Our solution can reduce time for finding joinable attributes without having to perform a manual data exploration on multiple datasets
