Abstract:Machine Learning (ML) has become pivotal across various fields, offering innovative solutions to complex data challenges. Professionals typically seek models that excel in both performance and reliability, aiming to achieve optimal generalization on future data. Since, then a variety of methods such as dummy coding, up/down-sampling, and bin-counting have been explored. However, finding a solution that effectively navigates the intricacies of limited and complex datasets still remains a challenge. This study introduces the K-Means Featurizer (KMF), an innovative algorithm crafted to enhance model performance and reliability, especially in scenarios involving complex and limited datasets. KMF employs K-Means clustering to generate enriched features that provide a nuanced understanding of the data, effectively balancing the similarity between the target variable and the feature space. This results in a more efficient predictive task by minimizing Euclidean distances and enhancing model generalizability. Our research validates KMF's effectiveness through an experiment in geoscience engineering, focusing on hydraulic conductivity (K) prediction, a vital parameter in well monitoring and infrastructure planning. Traditionally, K extraction is laborious and costly, requiring extensive pumping tests. KMF's application in this context demonstrates its potential to substantially reduce data losses during such operations. Applying KMF to the Extreme Gradient Boosting, Random Forest, K-Neighbors, Support Vector Machines, and Multiple Layers Neural Networks resulted in a significant improvement in prediction accuracy, with K-scores reaching up to 90%. While our experiment centers on geoscience engineering, KMF's utility extends to various domains facing similar data intricacies. Its adaptability to different types of complex datasets positions it as a valuable tool for diverse data-driven applications.

Multi-dimensional classification via kNN feature augmentation

Maximum Margin Multi-Dimensional Classification

Multi-Dimensional Classification via Decomposed Label Encoding

Probability-based Label Enhancement for Multi-Dimensional Classification

K-Ns: A Classifier by the Distance to the Nearest Subspace.

A General Multiple Data Augmentation Based Framework for Training Deep Neural Networks

Multi-dimensional Scaling from K-Nearest Neighbourhood Distances

Data-Dependent Kernel Discriminant Analysis for Feature Extraction and Classification

Combining KNN algorithm and other classifiers.

Multiple graph kernel learning based on GMDH-type neural network

Bayesian Model Selection Methods for Mutual and Symmetric $k$-Nearest Neighbor Classification

Incorporating Label Embedding and Feature Augmentation for Multi-Dimensional Classification.

Revisiting multi-dimensional classification from a dimension-wise perspective

DWS-MKL: Depth-width-scaling multiple kernel learning for data classification

K-Means Featurizer: A booster for intricate datasets

Exploiting Multi-Channels Deep Convolutional Neural Networks for Multivariate Time Series Classification

Virtual Classification: Modulating Domain-Specific Knowledge for Multidomain Crowd Counting

KNN Ensembles for Tweedie Regression: The Power of Multiscale Neighborhoods

Multi-Scale Dynamic Convolutional Network for Knowledge Graph Embedding

MK-MSMCR: An efficient multiple kernel approach to multi-class classification

Enhancing Peak Assignment in 13C NMR Spectroscopy: A Novel Approach Using Multimodal Alignment