Low Dimensional Fragment-Based Descriptors for Property Predictions in Inorganic Materials with Machine Learning

Aug 5, 2026

Upload by Liam

Machine learning accelerates inorganic materials discovery by replacing laborious trial-and-error characterization with data-driven prediction. This work introduces Low Dimensional Fragment Descriptors (LDFD), a simple fragment-based representation combined with ML models to predict key properties across diverse inorganic systems, including perovskites, alloys, semiconductors, and interfaces. Descriptor generation uses only structural formulas, applies binary encoding and dimensionality reduction, and enables efficient implementation. Evaluation on six datasets using up to eight components shows prediction quality comparable to prior methods, and suggests extensibility to layered or multi-site crystal systems with improved performance as unbiased data grows.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages22

Reading Time55 min

AccessFree to download

Introduction

Data-driven machine learning in materials science

Importance and challenges of descriptors

LDFD approach

Descriptor generation process

Applicability to inorganic systems

Performance evaluation

Dataset comparison

Targeted property predictions

Implementation and tools

Use of matminer

Expected outcomes and extensibility

Accuracy dependence on training data

Scalability with larger unbiased datasets

Low dimensional fragment-based descriptors for property predictions in inorganic materials with machine learning * Md Mohaiminul Islam , *Department of Mechanical Engineering, Temple University, Philadelphia, PA 19122, United States *[E-mail:](E-mail: md.mohaiminul.islam@temple.edu)[ ](E-mail: md.mohaiminul.islam@temple.edu)[md.mohaiminul.islam@temple.edu](E-mail: md.mohaiminul.islam@temple.edu) Abstract In recent times, the use of machine learning in materials design and discovery has aided to accelerate the discovery of innovative materials with extraordinary properties, which otherwise would have been driven by a laborious and time-consuming trial-and-error process. In this study, a simple yet powerful fragmentbased descriptor, Low Dimensional Fragment Descriptors (LDFD), is proposed to work in conjunction with machine learning models to predict important properties of a wide range of inorganic materials such asperovskite oxides, metal halide perovskites, alloys, semiconductor, and other materials system and can also be extended to work with interfaces. To predict properties, the generation of descriptors requires only the structural formula of the materials and, in presence of identical structure in the dataset, additional system properties as input. And the generation of descriptors involves few steps, encoding the formula in binary space and reduction of dimensionality, allowing easy implementation and prediction. To evaluate descriptor performance, six known datasets with up to eight components were compared. The method was applied to properties such as band gaps of perovskites and semiconductors, lattice constant of magnetic alloys, bulk/shear modulus of superhard alloys, critical temperature of superconductors, formation enthalpy and energy above hull convex of perovskite oxides. An advanced python-based data mining tool matminer was utilized for the collection of data. The prediction accuracies are equivalent to the quality of the training data and show comparable effectiveness as previous studies. This method should be extendable to any inorganic material systems which can be subdivided into layers or crystal structures with more than one atom site, and with the progress of data mining the performance should get better with larger and unbiased datasets. 1. Introduction Recently the material science community has experienced a growing interest in a data-driven predictive study with machine learning (ML) . Previously, advancements in material science have been serendipitous and slow. With the advent of data-driven ML methods, the material science community has incorporated ML methods in their workflow and greatly benefitted from data-driven approaches in terms of computational cost and time. At the heart of ML models lies data. The performance of ML, as a researcher, improves with training and large datasets. As ML predictive models become more and more popular and robust, researchers emphasized collecting data to be used in ML. To solve the issue with data more and more databases are being developed, data are being collected from computational and experimental studies ML models have enjoyed great success in predicting properties of materials i.e., band gap 1–3, critical temperature of superconductor4, shear, and bulk modulus5, Debye temperature6 and many more7,8 . One of the major advantages of ML models is that the prediction takes place in a very short time. Standard material characterization practices such as calculating the band structure are known to be notorious with Density functional theory (DFT) where calculations can take several days for prediction and going beyond standard DFT with random phase approximation or GW approximation9, the calculations can become so expensive computationally that it can become entirely impractical. Whereas ML methods offer very fast prediction with comparable accuracy with the help of previously calculated properties from data, as a result, the ML method

FAQs

What is the Low Dimensional Fragment Descriptors (LDFD) method used for?

LDFD is a fragment-based descriptor representation designed to work with machine learning models to predict important properties of inorganic materials, including systems that can be extended to interfaces.

What inputs are required to generate LDFD descriptors?

Descriptor generation requires only the structural formula of the materials; when identical structures exist in the dataset, additional system properties can be provided as input.

How does the study evaluate whether LDFD performs well?

Performance is evaluated by comparing descriptor effectiveness across six datasets (up to eight components) for properties such as band gaps, lattice constants, elastic moduli, superconducting critical temperatures, and formation-related energetics.

Aug 5, 2026

Upload by Liam