PHD Discussions Logo

Ask, Learn and Accelerate in your PhD Research

Question Icon Post Your Answer

Question Icon

How can models be identified from numerical data patterns?

In my lab, we generate complex datasets, and the first challenge is moving from the plotted curve to a candidate model. Beyond simple regression, what is the disciplined, often iterative, process experts use to hypothesize the governing equations from pattern recognition alone?

All Answers (1 Answers In All)

By Shreya K Answered 1 year ago

I would recommend framing this as a dialogue with your data, not a one-step extraction. First, you visually and statistically identify patterns is it exponential decay, oscillatory, logistic growth? This forms your hypothesis. Then, you select a model family (e.g., a differential equation or a specific distribution) that structurally produces such patterns. Using optimization for parameter estimation comes next. Crucially, I have seen the best researchers then validate with out-of-sample data and residual analysis. The model is only "identified" if it not only fits but also predicts and explains. It's an iterative cycle of conjecture and refutation.

Your Answer