← Back to home
The Foundation · Research

Research.

Published and in-progress work from ORIA contributors — spanning machine learning, renewable energy, additive manufacturing, and education research.

3 Published · 2 In Review
Published Work / 01

Published.

Three papers authored and co-authored by ORIA contributors, now published. Full text is available for each — click any paper to open it.

Paper 01 · Renewable Energy & Machine Learning Published

Modeling the Impact of Partial Shading on Photovoltaic Panels

AuthorsJ. Olivares et al.
AffiliationCSU Fullerton
VenueSpringer Nature
FieldMachine Learning
Abstract

Partial shading is a critical challenge for photovoltaic (PV) systems, causing severe power loss and damaging hotspots, yet existing detection approaches either rely on expensive hardware sensors or oversimplified analytical models that fail under dynamic conditions. This work presents an end-to-end machine learning approach for predicting shading percentage in PV systems using regression.

A custom MATLAB Simulink model implementing a physics-based single-diode equivalent circuit was used to generate a large-scale dataset of more than 24 million data points, systematically varying irradiance (200–1000 W/m²), temperature (10–50 °C), twelve series-parallel array configurations, and shading levels (0–100%). A context-aware feature-engineering step normalizes voltage, current, and power into scale-invariant ratios, enabling a single model to generalize across diverse panel configurations. The resulting XGBoost regression model achieves an R² of 0.937 with a mean absolute error of 5.19% and an RMSE of 7.76%, demonstrating that properly processed I–V curve data alone can accurately quantify shading without costly specialized hardware — a practical tool for optimizing residential PV systems.

Photovoltaics Partial shading XGBoost Regression Simulation dataset
Read paper (PDF)
Paper 02 · AI & Additive Manufacturing Published

Innovative Applications of AI and 3D Printing in Digital Dentistry: Enhancing Accuracy and Efficiency in Dental Care

AuthorsG. Martinez, K. Huang, Y. Lou, Y. Bai
AffiliationElectrical & Computer Eng., CSU Fullerton
VenueIEEE conference paper
FieldAI · 3D Printing
Abstract

Digital dentistry has significantly transformed the way dental professionals deliver patient care by integrating digital technologies into various aspects of the field, such as diagnosis, treatment planning, and restoration. This includes a broad array of technologies like computer-aided design and manufacturing (CAD/CAM), 3D printing, and artificial intelligence (AI), all of which are rapidly advancing and reshaping dental practice. While 3D printing has transformed industries, including dentistry, it still has several drawbacks and limitations — namely a complex design process, reliance on manual design adjustments, and limited resolution with occasional inaccuracies in 3D prints.

To address these issues, we propose an AI-enabled framework that provides a simplified process to ensure success and minimize human error. Users can quickly import an STL file into the software, select the desired application, and begin printing within seconds; the system automatically analyzes the 3D object, adjusting its orientation and generating supports for splints, crowns, bridges, and surgical guides. We focus on how AI can enhance the precision, efficiency, and reliability of dental prosthetics and restorations by automating critical steps that were traditionally manual and prone to errors — automatically detecting complex geometries, optimizing support structures, and adjusting printing parameters in real time. This AI-enabled approach reduces the number of design iterations while ensuring higher resolution and accuracy, overcoming the limitations of conventional 3D printing methods.

3D printing Digital dentistry Autoencoder Support optimization CAD/CAM
Read paper (PDF)
Paper 03 · Machine Learning & Education Published

Study on the Use of Random Forest Classifier and Multi-Output Classifier Models for Predicting Student Academic Performance and Identifying Areas of Concern

AuthorsK. Huang, I. Zimmerman, D. Bein
AffiliationTroy High School · CSU Fullerton
VenueASEE 2025
FieldMachine Learning
Abstract

This paper explores the use of machine learning to identify key factors that may connect to a student's academic performance, and how it may be used to predict student learning outcomes at an early stage — specifically, by utilizing two machine learning models: the Random Forest classifier and the Multi-Output classifier. The Random Forest Classifier operates by constructing multiple decision trees during training and selecting the mode of their predictions for a given input, identifying the most significant factors affecting outcomes. A Multi-Output Classifier is designed for multi-label tasks where each instance can be linked to multiple output variables, predicting several targets simultaneously — for example, assessing a student's grade and engagement level at once — and our implementation uses a neural network backend.

Several datasets sourced from Kaggle containing student background information and academic engagement and performance data were processed using the two classifier models, with the steps for cleaning, preparing, and analyzing the data discussed in the paper. The results show the Random Forest classifier is very effective at identifying key factors connected to academic performance — such as units completed in the previous semester, prior-semester grades, and tuition fee payment status — with an accuracy of 85.9% on the test data (94.5% correct on non-dropouts and 67.9% correct on dropouts). The same data processed by the Multi-Output Classifier neural network yielded accuracy scores ranging from 83.5% to 94.2% across the five target variables, providing valuable insights to educators for advocating tailored support for at-risk students.

Random forest Neural networks Multi-output classifier Early-warning systems Education
Read paper (PDF)
Under Review / 02

In review.

Active research not yet published. These are listed for completeness and will be linked here once they complete peer review.

OR · 02 · Causal Inference In review

Causal Structure Recovery under Partial Observability

AuthorsORIA Research
SubmissionNeurIPS '26 · Workshop
FieldState-space · Time series
StatusUnder peer review

We show that the standard estimate-then-discover pipeline for causal inference in partially observed dynamical systems is systematically biased when built on RTS-smoothed states. Filtered state estimates preserve the exogeneity conditions required for Granger-style discovery, and switching representation — at zero additional cost — recovers near-oracle causal structure.

Workshop submission currently under review. It will be linked here once the paper is published.
OR · 01 · Pedagogical AI In submission

Chalkless: Recovering the Logical Structure of Lectures

AuthorsORIA Research
TrackPedagogical AI
FieldEducation Technology
StatusSubmission in preparation

The research strand of the Chalkless initiative — a system for recovering the logical flow of a lecture: how definitions, examples, and key concepts develop over time, not just what is said. The work combines multimodal capture with sequence modeling to reconstruct the reasoning arc of a lecture in real time.

Paper submission in preparation. It will be linked here once published.