Neural Feature Learning From Relational Database
This addresses the tedious task of feature engineering for data scientists, showing a novel automated system that can compete effectively in real-world competitions.
The paper tackles the problem of automating feature learning from relational databases for predictive tasks, proving it is NP-hard and proposing an efficient rule-based and deep neural network approach that wins medals in Kaggle competitions and beats state-of-the-art solutions with significant margins.
Feature engineering is one of the most important but most tedious tasks in data science. This work studies automation of feature learning from relational database. We first prove theoretically that finding the optimal features from relational data for predictive tasks is NP-hard. We propose an efficient rule-based approach based on heuristics and a deep neural network to automatically learn appropriate features from relational data. We benchmark our approaches in ensembles in past Kaggle competitions. Our new approach wins late medals and beats the state-of-the-art solutions with significant margins. To the best of our knowledge, this is the first time an automated data science system could win medals in Kaggle competitions with complex relational database.