Selected work

Representative Papers

A novel association and ranking approach identifies factors affecting educational outcomes of STEM majors

Mar 16, 2025

This study identifies actionable, intervention-sensitive factors influencing undergraduate STEM graduation rates to enable evidence-based policy design. Method: Leveraging integrated administrative data—including academic transcripts, demographic attributes, institutional records, and, for the first time, National Student Clearinghouse (NSC) transfer-tracking data—from two four-year institutions in the U.S. Northeast, we apply the D-basis formal concept analysis algorithm to uncover causal associations, explicitly incorporating post-transfer degree completion outcomes. Contribution/Results: We reveal a counterintuitive positive association between STEM-to-non-STEM major switching and higher overall graduation rates. Key predictive factors include introductory biology/chemistry/mathematics course performance, initial mathematics course difficulty selection, and institutional flexibility in major changes. Variables such as Pell Grant eligibility significantly reflect structural inequities in time-to-degree and retention. Findings provide empirically grounded, operationally actionable insights for targeted STEM education interventions.

0 citationsRead paper

Empowering Agricultural Insights: RiceLeafBD - A Novel Dataset and Optimal Model Selection for Rice Leaf Disease Diagnosis through Transfer Learning Technique

Jan 15, 2025

To address rice leaf disease–induced yield loss and food insecurity in densely populated, arable land–constrained regions like Bangladesh, this study introduces RiceLeafBD—the first high-diversity, annotation-bias-free, field-collected rice leaf disease image dataset from Bangladesh, specifically designed for major rice-producing countries. Leveraging RiceLeafBD, we conduct systematic comparative evaluations of lightweight transfer learning models (e.g., MobileNet-V2, EfficientNet-V2), establishing a model selection paradigm tailored to resource-constrained agricultural settings. Experimental results demonstrate that EfficientNet-V2 achieves 91.5% classification accuracy—surpassing existing state-of-the-art methods—and confirms both dataset validity and feasibility of edge deployment. Our core contributions are twofold: (1) the release of the first real-world, field-acquired benchmark dataset for rice disease diagnosis; and (2) a lightweight diagnostic framework that jointly optimizes accuracy and computational efficiency for on-farm deployment.

0 citationsRead paper

Recent publications

Latest Papers

A novel association and ranking approach identifies factors affecting educational outcomes of STEM majors

Mar 16, 2025

This study identifies actionable, intervention-sensitive factors influencing undergraduate STEM graduation rates to enable evidence-based policy design. Method: Leveraging integrated administrative data—including academic transcripts, demographic attributes, institutional records, and, for the first time, National Student Clearinghouse (NSC) transfer-tracking data—from two four-year institutions in the U.S. Northeast, we apply the D-basis formal concept analysis algorithm to uncover causal associations, explicitly incorporating post-transfer degree completion outcomes. Contribution/Results: We reveal a counterintuitive positive association between STEM-to-non-STEM major switching and higher overall graduation rates. Key predictive factors include introductory biology/chemistry/mathematics course performance, initial mathematics course difficulty selection, and institutional flexibility in major changes. Variables such as Pell Grant eligibility significantly reflect structural inequities in time-to-degree and retention. Findings provide empirically grounded, operationally actionable insights for targeted STEM education interventions.

0 citationsRead paper

Empowering Agricultural Insights: RiceLeafBD - A Novel Dataset and Optimal Model Selection for Rice Leaf Disease Diagnosis through Transfer Learning Technique

Jan 15, 2025

To address rice leaf disease–induced yield loss and food insecurity in densely populated, arable land–constrained regions like Bangladesh, this study introduces RiceLeafBD—the first high-diversity, annotation-bias-free, field-collected rice leaf disease image dataset from Bangladesh, specifically designed for major rice-producing countries. Leveraging RiceLeafBD, we conduct systematic comparative evaluations of lightweight transfer learning models (e.g., MobileNet-V2, EfficientNet-V2), establishing a model selection paradigm tailored to resource-constrained agricultural settings. Experimental results demonstrate that EfficientNet-V2 achieves 91.5% classification accuracy—surpassing existing state-of-the-art methods—and confirms both dataset validity and feasibility of edge deployment. Our core contributions are twofold: (1) the release of the first real-world, field-acquired benchmark dataset for rice disease diagnosis; and (2) a lightweight diagnostic framework that jointly optimizes accuracy and computational efficiency for on-farm deployment.

0 citationsRead paper