Nanostructures, defined as materials with at least one dimension smaller than 100 nm, have

fundamentally transformed material science by unlocking novel physicochemical properties.

Based on their morphology, they are categorized as zero-dimensional (0D), one-dimensional
(1D), two-dimensional (2D), or three-dimensional (3D), each offering unique advantages across
diverse applications, including electronics, catalysis, and biomedicine. The remarkable impact of
nanostructures stems from their size-dependent properties, particularly in optical and catalytic
performance, which enable highly specialized functionalities.

However, accurately characterizing these materials is essential for tailoring their properties to
specific applications. Traditional techniques, such as electron microscopy, while precise, are
expensive, require sophisticated instrumentation, and can only be performed post-synthesis. This
limitation prevents real-time monitoring and optimization of synthesis conditions, restricting the
ability to fine-tune nanostructure properties dynamically.

To overcome these challenges, machine learning (ML) has emerged as a transformative tool in
nanomaterials research, offering an efficient and cost-effective alternative for property
prediction. ML algorithms can analyze complex datasets, extract meaningful patterns, and
predict nanoparticle properties with remarkable accuracy—potentially reducing reliance on
labor-intensive characterization techniques. Studies by Pellegrino et al. and Gloubitz et al. have
demonstrated ML’s effectiveness in predicting the sizes of titanium oxide and gold
nanoparticles, respectively. Similarly, Nyeo et al. employed ML models to derive particle size
distributions from dynamic light scattering (DLS) data, further highlighting its applicability.

Despite its promise, the application of ML in nanoparticle characterization faces a critical hurdle:
the scarcity of high-quality experimental datasets. To address this, Pashkov et al. simulated UV-
Vis spectra to train ML models for predicting gold nanoparticle size and shape. However, models
trained exclusively on simulated data often exhibit poor generalizability to real experimental
conditions due to systematic errors and inherent limitations in the mathematical models used.
This underscores the pressing need for experimentally derived datasets to fully harness ML’s
potential in advancing nanomaterials research. Unfortunately, well-structured and openly
accessible datasets for ML applications remain largely unavailable.
To bridge this gap, data digitization from previously published studies has emerged as a viable
strategy for compiling datasets suitable for ML applications. For instance, Kwaria et al.
successfully demonstrated the use of data extraction techniques to build datasets for predicting
protein adsorption on self-assembled monolayers. Inspired by this approach, our study focuses
on extracting and curating experimental data from published literature to construct a robust
dataset for ML-driven nanoparticle characterization.

However, data digitization introduces challenges such as class imbalance, where certain input
parameters are overrepresented while others are underrepresented, leading to biased predictions.
To mitigate this issue, careful pre-research is necessary to identify materials with extensive and
well-documented synthesis and characterization data. Zinc oxide (ZnO) nanoparticles were
selected for this study due to their well-documented experimental research history. ZnO exhibits
a wide bandgap (3.3 eV) and high exciton binding energy (60 meV), making it a technologically
significant material with applications in optoelectronics, photovoltaics, and biomedicine.

By systematically digitizing and structuring ZnO synthesis conditions and UV-Vis-derived

energy gap data, we developed an ML-based predictive model for ZnO nanoparticle size. This
approach underscores the transformative potential of integrating machine learning and data-
driven methodologies to revolutionize nanomaterials research and applications.

Nanostructures, defined as materials with dimensions smaller than 100 nm, have revolutionized
material science by revealing unique physicochemical properties. These structures are
categorized by their morphology into zero-dimensional (0D), one-dimensional (1D), two-
dimensional (2D), or three-dimensional (3D) types. Each category offers distinct advantages
across various applications, such as electronics, catalysis, and biomedicine. The significant
impact of nanostructures is primarily due to their size-dependent properties, especially in optical
and catalytic performance, enabling specialized functionalities. Therefore, accurate
characterization of these materials is crucial for tailoring their properties to specific applications.
Traditional techniques, like electron microscopy, are precise but expensive, require sophisticated
instrumentation, and can only be performed post-synthesis. This limitation hinders real-time
monitoring and optimization of synthesis conditions, restricting the dynamic fine-tuning of
nanostructure properties.

Machine learning (ML) has emerged as a transformative tool in nanomaterials research, offering
an efficient and cost-effective alternative for property prediction. ML algorithms can analyze
complex datasets, extract meaningful patterns, and predict nanoparticle properties with
remarkable accuracy, potentially reducing reliance on labor-intensive characterization
techniques. Studies by Pellegrino et al. and Gloubitz et al. have shown ML’s effectiveness in
predicting the sizes of titanium oxide and gold nanoparticles, respectively. Similarly, Nyeo et al.
used ML models to derive particle size distributions from dynamic light scattering (DLS) data,
further highlighting its applicability.

Despite its promise, the application of ML in nanoparticle characterization faces a significant

challenge: the scarcity of high-quality experimental datasets. Pashkov et al. addressed this by
simulating UV-Vis spectra to train ML models for predicting gold nanoparticle size and shape.
However, models trained solely on simulated data often exhibit poor generalizability to real
experimental conditions due to systematic errors and inherent limitations in the mathematical
models used. This underscores the need for experimentally derived datasets to fully harness
ML’s potential in advancing nanomaterials research. Unfortunately, well-structured and openly
accessible datasets for ML applications remain largely unavailable.

To bridge this gap, data digitization from previously published studies has emerged as a viable
strategy for compiling datasets suitable for ML applications. For example, Kwaria et al.
successfully demonstrated the use of data extraction techniques to build datasets for predicting
protein adsorption on self-assembled monolayers. Inspired by this approach, our study focuses
on extracting and curating experimental data from published literature to construct a robust
dataset for ML-driven nanoparticle characterization.

However, data digitization introduces challenges such as class imbalance, where certain input
parameters are overrepresented while others are underrepresented, leading to biased predictions.
To mitigate this issue, careful pre-research is necessary to identify materials with extensive and
well-documented synthesis and characterization data. Zinc oxide (ZnO) nanoparticles were
selected for this study due to their well-documented experimental research history. ZnO exhibits
a wide bandgap (3.3 eV) and high exciton binding energy (60 meV), making it a technologically
significant material with applications in optoelectronics, photovoltaics, and biomedicine.

By systematically digitizing and structuring ZnO synthesis conditions and UV-Vis-derived

energy gap data, we developed an ML-based predictive model for ZnO nanoparticle size. This
approach highlights the transformative potential of integrating machine learning and data-driven
methodologies to revolutionize nanomaterials research and applications.

Nanostructures, materials with at least one dimension smaller than 100 nm, have revolutionized
modern science by unlocking unique physicochemical properties. Depending on their shape, they
can be classified as 0D, 1D, 2D, or 3D, each offering distinct advantages in fields ranging from
electronics to medicine. Their true potential became evident when researchers discovered that
size could dramatically influence a material’s optical and catalytic behavior, paving the way for
groundbreaking applications.

However, harnessing these properties requires precise characterization, a challenge often

hindered by traditional techniques like electron microscopy. While highly effective, these
methods are expensive, require specialized equipment, and are only applicable post-synthesis—
offering no real-time insights or opportunities to fine-tune synthesis conditions.

To bridge this gap, machine learning has emerged as a game-changer in nanomaterials research.
By leveraging vast datasets, machine learning models can predict nanoparticle properties with
remarkable accuracy, potentially replacing costly and time-consuming characterization
techniques. Researchers such as Pellegrino et al. and Gloubitz et al. have already demonstrated
the power of machine learning by successfully predicting the sizes of titanium oxide and gold
nanoparticles. As this field continues to evolve, machine learning promises to revolutionize
nanomaterials development, making precise, real-time characterization more accessible than ever
before.

Nanostructures represent a fascinating frontier in materials science, defined by their incredibly

small scale—with at least one dimension measuring less than 100 nanometers. These materials
are categorized into 0D, 1D, 2D, or 3D based on their structural shapes, each offering unique
properties and potential applications. What makes nanostructures truly remarkable is the
discovery that their size can dramatically influence their physicochemical characteristics, such as
optical behavior and catalytic efficiency. This size-dependent versatility has opened the door to
groundbreaking applications across various fields.

However, unlocking the full potential of nanomaterials hinges on precise characterization.

Traditional methods, like electron microscopy, while effective, come with significant drawbacks:
they are often prohibitively expensive and inaccessible to many research labs. Moreover, these
techniques are typically employed after synthesis, leaving scientists in the dark about
nanoparticle size and optical properties during the synthesis process. This limitation prevents
real-time adjustments, hindering optimization efforts.

Enter machine learning—a game-changing solution to this challenge. By leveraging advanced

algorithms, researchers can now predict nanoparticle properties with remarkable accuracy,
potentially replacing costly and time-intensive characterization methods. Pioneering work by
Pellegrino et al. and Gloubitz et al. has already demonstrated the power of machine learning in
predicting the sizes of titanium oxide and gold nanoparticles, respectively. This innovative
approach not only saves time and resources but also paves the way for real-time synthesis
optimization, revolutionizing the field of nanotechnology.