Measure Your Models Honestly

·1 min read·

I built a credit scoring model expecting brilliance. The metrics told the truth. On measuring ML model quality honestly.

When I trained the first credit scoring model for AfriCred, I was confident. The model underperformed my expectations — until I benchmarked it against commercial credit bureaus in comparable markets. In that context, it was a significant win.

The lesson: measure honestly. Do not round up. Do not explain away the gaps. Accept what the metrics say, benchmark against the right baseline, then decide.

My gut was both overconfident and unable to benchmark without data.

Models do not have feelings. They have numbers. Read the numbers — then benchmark them.

— Dick Bassey | DevDick | 2019

Share this article
← Back to all writing