MinLinMo: A Minimalist Approach to Variable Selection and Linear Model Prediction
Author Information
Author(s): Bohlin Jon, Håberg Siri E., Magnus Per, Gjessing Håkon K.
Primary Institution: Norwegian Institute of Public Health
Hypothesis
Can a minimalist software approach improve variable selection and prediction accuracy in high-dimensional datasets?
Conclusion
MinLinMo produces smaller, parsimonious models that maintain acceptable prediction accuracy compared to larger models like Elastic Net.
Supporting Evidence
- MinLinMo models perform comparably to established prediction models requiring hundreds of predictors.
- MinLinMo was trained on the Norwegian Mother, Father and Child Birth Cohort datasets.
- MinLinMo selected fewer predictors while maintaining acceptable prediction accuracy.
Takeaway
MinLinMo is a tool that helps scientists make predictions using fewer variables, making it easier to understand the results.
Methodology
MinLinMo uses Pearson correlations to select variables and builds models based on those that improve prediction accuracy.
Potential Biases
Potential for missing predictors due to the priority queue selection method.
Limitations
MinLinMo can only work with linear models and may miss complex correlation structures.
Participant Demographics
The study included 2,188 samples, with 51.5% females and 48.5% males.
Statistical Information
P-Value
p<0.05
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website