Machine Learning Analysis of Dimensionality Effects on Bandgap Prediction

Bandgap engineering is central to tailoring the electronic and optical properties of semiconductors and low-dimensional materials. In this work, we present a machine learning (ML) framework capable of accurately predicting key material properties such as formation energy, energy above the hull, and bandgap across various dimensionalities. Unlike conventional approaches, our model explicitly incorporates quantum confinement effects by integrating structural descriptors and dimensionality-based features, enabling systematic insights into confinement-induced bandgap variations. Our data set, curated from a diverse set of materials, includes atomic composition, crystallographic symmetry, and structural features but excludes DFT or experimentally derived property data such as the information on stability, bandgap, and their nature. To compensate, we developed predictive models trained on external property references, achieving strong performance with an XGBRegressor yielding an R2 score of 0.875. A k-fold cross-validation strategy ensures robustness and generalizability, while Optuna-based hyperparameter tuning minimizes overfitting and optimizes predictive accuracy. Analysis reveals that bandgap trends are strongly influenced by average ionic character and dimensionality, with 0D materials exhibiting the most pronounced quantum confinement effects. To demonstrate practical utility, we applied our model to the crystallography materials repository (CMR), identifying promising candidates for light-harvesting applications. Selected materials were further validated using density functional theory (DFT) calculations to confirm structural stability and electronic properties. This work provides a scalable pathway for high-throughput discovery and bandgap prediction in low-dimensional materials.