Accuracy, Efficiency, and Class-Wise Uncertainty in Dry-Bean Classification: A Reproducible Case Study

This preprint presents a fully reproducible case study of accuracy, inference efficiency, and class-wise uncertainty on the UCI Dry Bean classification benchmark. It evaluates a calibrated elastic-net logistic-regression model against CPU baselines under a duplicate-safe split protocol and uses global and class-wise split-conformal prediction sets. The proposed compact model did not meet its pre-specified accuracy threshold relative to the strongest nonlinear baseline; this negative trade-off is reported transparently. This record is a preprint and has not been peer reviewed. Source code, fixed split metadata, and reproduction instructions:https://github.com/anishreddy13/ml_research Dataset:UCI Dry Bean Dataset (ID 602)https://archive.ics.uci.edu/dataset/602/dry%2Bbean%2Bdataset Code and experiment infrastructure were developed with AI coding assistance; all results, analysis, and conclusions were reviewed and are the sole responsibility of the author.