-
Methodology to Create Analysis-Naive Holdout Records as well as Train and Test Records for Machine Learning Analyses in Healthcare
Authors:
Michele Bennett,
Mehdi Nekouei,
Armand Prieditis Rajesh Mehta,
Ewa Kleczyk,
Karin Hayes
Abstract:
It is common for researchers to holdout data from a study pool to be used for external validation as well as for future research, and the same desire is true to those using machine learning modeling research. For this discussion, the purpose of the holdout sample it is preserve data for research studies that will be analysis-naive and randomly selected from the full dataset. Analysis-naive are rec…
▽ More
It is common for researchers to holdout data from a study pool to be used for external validation as well as for future research, and the same desire is true to those using machine learning modeling research. For this discussion, the purpose of the holdout sample it is preserve data for research studies that will be analysis-naive and randomly selected from the full dataset. Analysis-naive are records that are not used for testing or training machine learning (ML) models and records that do not participate in any aspect of the current machine learning study. The methodology suggested for creating holdouts is a modification of k-fold cross validation, which takes into account randomization and efficiently allows a three-way split (holdout, test and training) as part of the method without forcing. The paper also provides a working example using set of automated functions in Python and some scenarios for applicability in healthcare.
△ Less
Submitted 8 May, 2022;
originally announced May 2022.
-
Universal mean moment rate profiles of earthquake ruptures
Authors:
Amit P. Mehta,
Karin A. Dahmen,
Yehuda Ben-Zion
Abstract:
Earthquake phenomenology exhibits a number of power law distributions including the Gutenberg-Richter frequency-size statistics and the Omori law for aftershock decay rates. In search for a basic model that renders correct predictions on long spatio-temporal scales, we discuss results associated with a heterogeneous fault with long range stress-transfer interactions. To better understand earthqu…
▽ More
Earthquake phenomenology exhibits a number of power law distributions including the Gutenberg-Richter frequency-size statistics and the Omori law for aftershock decay rates. In search for a basic model that renders correct predictions on long spatio-temporal scales, we discuss results associated with a heterogeneous fault with long range stress-transfer interactions. To better understand earthquake dynamics we focus on faults with Gutenberg-Richter like earthquake statistics and develop two universal scaling functions as a stronger test of the theory against observations than mere scaling exponents that have large error bars. Universal shape profiles contain crucial information on the underlying dynamics in a variety of systems. As in magnetic systems, we find that our analysis for earthquakes provides a good overall agreement between theory and observations, but with a potential discrepancy in one particular universal scaling function for moment-rates. The results reveal interesting connections between the physics of vastly different systems with avalanche noise.
△ Less
Submitted 9 September, 2005;
originally announced September 2005.
-
Higher Order Spectra in Avalanche Models
Authors:
Amit P. Mehta,
Karin A. Dahmen,
A. C. Mills,
M. B. Weissman
Abstract:
Utilizing the Haar transform, we study the higher order spectral properties of mean field avalanche models, whose avalanche dynamics are described by Poisson statistics at a critical point or critical depinning transition. The Haar transform allows us to obtain a time series of noise powers, $H(f_1,t)$, that gives improved time resolution over the Fourier transform. Using $H(f_1,t)$ we analytica…
▽ More
Utilizing the Haar transform, we study the higher order spectral properties of mean field avalanche models, whose avalanche dynamics are described by Poisson statistics at a critical point or critical depinning transition. The Haar transform allows us to obtain a time series of noise powers, $H(f_1,t)$, that gives improved time resolution over the Fourier transform. Using $H(f_1,t)$ we analytically calculate the Haar power spectrum, the real 1.5 spectra, the second spectra, and the real cross second spectra in mean field avalanche models. We verify our theoretical results with the numerical results from a simulation of the T=0 mean field nonequilibrium random field Ising model (RFIM). We also extend our higher order spectra calculation to data obtained from a numerical simulation of the T=0 infinite range RFIM for $d=3$, and experimental data obtained from an amorphous alloy, $Fe_{21}Co_{64}B_{15}$. We compare the results and obtain novel exponents.
△ Less
Submitted 2 March, 2005;
originally announced March 2005.
-
Noise in Disordered Systems: Higher Order Spectra in Avalanche Models
Authors:
Amit P. Mehta,
Karin A. Dahmen,
M. W. Weissman,
Tim Wotherspoon
Abstract:
We present a novel analytic calculation of the Haar power spectra, and various higher order spectra, of mean field avalanche models. We also compute these spectra from a simulation of the zero-temperature mean field RFIM and infinite range RFIM model for $d = 3$. We compare the results and obtain novel exponents.
We present a novel analytic calculation of the Haar power spectra, and various higher order spectra, of mean field avalanche models. We also compute these spectra from a simulation of the zero-temperature mean field RFIM and infinite range RFIM model for $d = 3$. We compare the results and obtain novel exponents.
△ Less
Submitted 25 January, 2005;
originally announced January 2005.
-
Universal shape profiles of earthquake ruptures
Authors:
Amit P. Mehta,
Karin A. Dahmen,
Yehuda Ben-Zion
Abstract:
Universal shape profiles in a variety of systems contain crucial information on the underlying dynamics. We develop such shape profiles for earthquakes as a stronger test of theory against observations. The earthquake analysis shows good agreement between theory and observations, with a discrepancy in one universal profile, just as in magnets. The analysis provides a stronger support for the ide…
▽ More
Universal shape profiles in a variety of systems contain crucial information on the underlying dynamics. We develop such shape profiles for earthquakes as a stronger test of theory against observations. The earthquake analysis shows good agreement between theory and observations, with a discrepancy in one universal profile, just as in magnets. The analysis provides a stronger support for the idea that earthquakes are associated with criticality rather than event statistics only. The results point to the existence of deep connections between the physics of avalanches in different systems.
△ Less
Submitted 22 March, 2004;
originally announced March 2004.
-
Universal Pulse Shape Scaling Function and Exponents: A Critical Test for Avalanche Models applied to Barkhausen Noise
Authors:
Amit P. Mehta,
Andrea C. Mills,
Karin Dahmen,
James P. Sethna
Abstract:
In order to test if the universal aspects of Barkhausen noise in magnetic materials can be predicted from recent variants of the non-equilibrium zero temperature Random Field Ising Model (RFIM), we perform a quantitative study of the universal scaling function derived from the
Barkhausen pulse shape in simulations and experiment. Through data collapses and scaling relations we determine the cr…
▽ More
In order to test if the universal aspects of Barkhausen noise in magnetic materials can be predicted from recent variants of the non-equilibrium zero temperature Random Field Ising Model (RFIM), we perform a quantitative study of the universal scaling function derived from the
Barkhausen pulse shape in simulations and experiment. Through data collapses and scaling relations we determine the critical exponents $τ$ and $1/σνz$ in both simulation and experiment. Although we find agreement in the critical exponents, we find differences between theoretical and experimental pulse shape scaling functions as well as between different experiments.
△ Less
Submitted 7 December, 2001; v1 submitted 6 December, 2001;
originally announced December 2001.
-
Topological Invariants in Microscopic Transport on Rough Landscapes: Morphology, Hierarchical Structure, and Horton Analysis of Riverlike Networks of Vortices
Authors:
A. P. Mehta,
C. Reichhardt,
C. J. Olson,
Franco Nori
Abstract:
River basins as diverse as the Nile, the Amazon, and the Mississippi satisfy certain topological invariants known as Horton's laws. Do these macroscopic (up to 10^3 km) laws extend to the micron scale? Through realistic simulations, we analyze the morphology and statistical properties of networks of vortex flow in flux-gradient-driven superconductors. We derive a phase diagram of the different n…
▽ More
River basins as diverse as the Nile, the Amazon, and the Mississippi satisfy certain topological invariants known as Horton's laws. Do these macroscopic (up to 10^3 km) laws extend to the micron scale? Through realistic simulations, we analyze the morphology and statistical properties of networks of vortex flow in flux-gradient-driven superconductors. We derive a phase diagram of the different network morphologies, including one in which Horton's laws of length and stream number are obeyed -- even though these networks are about 10^9 to 10^15 times smaller than geophysical river basins.
△ Less
Submitted 28 June, 1999;
originally announced June 1999.
