Detecting anti-patterns in SQL Queries using Text Classification Techniques

A major problem with using relational databases, is writing efficient SQL queries. Some common errors known as anti-patterns are frequent in SQL queries and can seriously impact the query execution time and sometimes, the database general performance. This paper shows how ma-chine learning techniques can be lever-aged to detect anti-patterns in SQL queries by approaching the problem as a text classification problem. Our result is a model based on a convolutional neural net-work that can be used to classify a SQL query into zero, one or many anti-patterns classes.