Enhanced Cross-Validation Methods Leveraging Clustering Techniques
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
The efficacy of emerging and established learning algorithms warrants scrutiny.This examination is intrinsically linked to the results of classification performance.The primary determinant influencing these results is the distribution of the training and test data presented to the algorithms.Existing literature frequently employs standard and stratified (S-CV and St-CV) k-fold cross-validation methods for the creation of training and test data for classification tasks.In the S-CV method, training and test groups are formed via random data distribution, potentially undermining the reliability of performance results calculated post-classification.This study introduces innovative cross-validation strategies based on kmeans and k-medoids clustering to address this challenge.These strategies are designed to tackle issues emerging from random data distribution.The proposed methods autonomously determine the number of clusters and folds.Initially, the number of clusters is established via Silhouette analysis, followed by identifying the number of folds according to the data volume within these clusters.An additional aim of this study is to minimize the standard deviation (Std) values between the folds.Particularly in classifying large datasets, the minimized Std negates the need to present each fold to the system, thereby reducing time expenditure and system congestion/fatigue.Analyses were carried out on several large-scale datasets to demonstrate the superiority of these new CV methods over the S-CV and St-CV techniques.The findings revealed superior performance results for the novel strategies.For instance, while the minimum Std value between folds was 0.022, the maximum accuracy rate achieved was approximately 100%.Owing to the proposed methods, the discrepancy between the performance outputs of each fold and the overall average is statistically minimized.The randomness in creating the training/test groups, which has been previously identified as a negative contributing factor to this discrepancy, has been significantly reduced.Hence, this study is anticipated to fill a critical and substantial gap in the existing literature concerning the formation of training/test groups in various classification problems and the statistical accuracy of performance results.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.002 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it