Empirical Analysis of the Adoption and Performance of AutoML Tools Among Nigerian Data Practitioners

Automated machine learning tools promise to democratise model development by reducing the technical barrier, yet their actual adoption rates, effective use, and limitations among Nigerian data science practitioners have not been empirically studied. This study empirically examined AutoML tool adoption, usage patterns, and model quality outcomes among data science practitioners in Nigeria. A mixed-methods design was employed: an online survey of 280 Nigerian data practitioners recruited through LinkedIn and local data science communities assessed tool awareness, adoption, workflow integration, and perceived quality, while 60 consenting practitioners submitted model performance results from a standardised benchmark task comparing their AutoML-produced models against manually tuned baselines. Survey results showed that AutoML tool awareness was high at 76.8 percent but active deployment was 34.3 percent. H2O AutoML, Auto-sklearn, and Google AutoML were the most commonly used tools. AutoML was most frequently used for rapid prototyping rather than production deployment. Benchmark analysis revealed that AutoML-produced models achieved 92.7 percent of the performance of manually tuned expert models on structured tabular data tasks, but 78.4 percent on time-series tasks and 71.3 percent on imbalanced class problems. Practitioners with fewer than two years of experience showed the highest AutoML adoption rates. The study fills an empirical adoption and performance gap for the Nigerian data science community and recommends AutoML integration into university data science curricula for foundational tasks, alongside professional training on its limitations for complex problem types.

Keywords: AutoML, machine learning automation, Nigerian data practitioners, model performance, adoption study