[Paper Review] Augmented Data Science: Towards Industrialization and Democratization of Data Science

Hüseyin Uzunalioğlu, Jin Cao+6 more|arXiv (Cornell University)|Sep 12, 2019

Scientific Computing and Data Management6 citations

TL;DR

This paper introduces Augmented Data Science (ADS), a domain-agnostic, end-to-end framework that automates data exploration, cleaning, feature engineering, and model selection using statistics and machine learning to reduce manual effort in data science. The approach achieves strong performance—AUC of 0.781 on a real-world dataset—demonstrating its potential to industrialize and democratize data science with minimal human intervention.

ABSTRACT

Conversion of raw data into insights and knowledge requires substantial amounts of effort from data scientists. Despite breathtaking advances in Machine Learning (ML) and Artificial Intelligence (AI), data scientists still spend the majority of their effort in understanding and then preparing the raw data for ML/AI. The effort is often manual and ad hoc, and requires some level of domain knowledge. The complexity of the effort increases dramatically when data diversity, both in form and context, increases. In this paper, we introduce our solution, Augmented Data Science (ADS), towards addressing this "human bottleneck" in creating value from diverse datasets. ADS is a data-driven approach and relies on statistics and ML to extract insights from any data set in a domain-agnostic way to facilitate the data science process. Key features of ADS are the replacement of rudimentary data exploration and processing steps with automation and the augmentation of data scientist judgment with automatically-generated insights. We present building blocks of our end-to-end solution and provide a case study to exemplify its capabilities.

Motivation & Objective

• To address the 'human bottleneck' in data science, where up to 80% of effort is spent on data preparation rather than modeling.
• To industrialize data science by enabling faster, scalable insight extraction from diverse datasets without deep domain expertise.
• To democratize data science by empowering non-experts to perform basic data science tasks through automation and intelligent augmentation.
• To develop a task-agnostic, robust, and interactive system that works with raw, low-quality data and supports user intervention when needed.

Proposed method

• ADS uses automated statistical and machine learning techniques to analyze data structure, quality, and relationships without requiring domain knowledge.
• It applies unsupervised and semi-supervised methods to detect data quality issues, such as missingness patterns, correlations, and anomalies.
• The system generates features automatically using a data-driven approach, with a focus on preserving data semantics and relationships.
• It employs model selection pipelines using algorithms like lightGBM and Bayesian optimization to identify the best-performing model.
• The framework supports user augmentation by providing visual and analytical insights, allowing experts to intervene when automation is insufficient.
• It is designed as an end-to-end pipeline that chains data discovery, feature engineering, and model tuning into a single, interactive workflow.

Experimental results

Research questions

• RQ1 Can automated, domain-agnostic techniques significantly reduce the manual effort required in data preparation and feature engineering?
• RQ2 How effective is an end-to-end, data-driven system in generating high-quality features and models without human domain expertise?
• RQ3 To what extent can automated insights augment human judgment in data science tasks, especially when data quality is poor or complex?
• RQ4 Can such a system achieve competitive performance compared to expert-optimized models with minimal human input?

Key findings

• ADS achieved an AUC of 0.781 on a real-world credit risk dataset using only automated feature generation and model selection, demonstrating strong baseline performance.
• The system successfully identified systematic missingness patterns—such as 73% of 31 columns missing together—enabling better data handling decisions.
• It detected highly correlated column pairs and structural patterns (e.g., cliques in association graphs), revealing hidden data relationships without prior assumptions.
• The framework reduced the need for manual data exploration and preprocessing, enabling rapid insight extraction from new and diverse datasets.
• Top-performing models on public leaderboards required hundreds of submissions and expert tuning, whereas ADS achieved near-top performance with just a few automated runs.
• The results suggest that automated feature engineering and model selection can produce competitive models even without hand-crafted domain features, though performance can be further improved by integrating expert knowledge.