Replacing Euclidean distance with random-forest weights, FORWS and FORWC match or beat conventional tools on noisy high-dimensional series and extract directed interactions between honeybee-hive acoustic vectors and scalar temperature

Introduces FORWS and FORWC, which substitute adaptive "forest weights" from random forest ensembles for the Euclidean distance metric in forecasting and causal inference for non-linear dynamical systems. Existing Empirical Dynamic Modeling reconstructs state space in Euclidean space, where distance measurements become distorted as dimensions grow; this work swaps that metric for data-driven forest weights. The abstract states validation on simulated and empirical ecological datasets and reports forecasting skill comparable to or better than conventional tools; specific datasets, sample sizes, and effect sizes are not given in the abstract.

FORWS/FORWC show substantial resilience against dynamic process noise and mitigate the curse of dimensionality. Noise robustness and high-dimensional adaptability are presented as core properties of the method rather than accuracy on a single dataset. The abstract uses the phrases "substantial resilience" and "mitigating the curse of dimensionality"; these are the author's summary conclusions from simulated and empirical data, with no quantitative comparison provided in the abstract.

FORWC enables multimodal causal inference, successfully extracting directed interactions between high-dimensional acoustic vectors and scalar temperature data monitored in a honeybee hive. Extends causal inference from homogeneous, low-dimensional variables to mixed multimodal settings combining high-dimensional vectors and scalars. The abstract provides the honeybee acoustic-and-temperature monitoring case as a concrete empirical example, but reports no causal strength, significance testing, or control setup.

The framework bridges chaotic physics and machine learning to extract intrinsic manifold topology, offering a paradigm for uncovering causal networks in complex real-world environments. Positions the work as a methodological bridge rather than a single-domain application result. This is a framework-level statement at the abstract level; its generality awaits testing on more domain data.

The work targets researchers who need both forecasting and causal inference from high-dimensional, noisy, non-linear time series, and it suits settings where state-space reconstruction suffers from the curse of dimensionality and where variable types are mixed (high-dimensional vectors alongside scalars), with honeybee acoustic and temperature monitoring as one such case. Because the method replaces the Euclidean metric with weights from random forest ensembles, its applicability presupposes enough observations to train the ensemble. For researchers hoping to extend causal-network analysis from low-dimensional homogeneous variables to multimodal sensor data, this framework offers a directly borrowable technical route.

The available text is only the abstract and the competing-interest statement; the main text, figures, and supplementary materials are not included, so the scale of the simulated and empirical datasets, the noise settings, the way comparisons with conventional tools were made, and the statistical support for causal direction in the honeybee case cannot be checked. The quantitative basis for phrases such as "comparable or improved forecasting skill" and "substantial resilience," and the conditions under which multimodal causal inference would fail, remain open questions answerable only by reading the full paper. In addition, the author declares a pending patent application related to FORWS/FORWC, which readers may treat as background when assessing method adoption.