Background/Context: Predictive analytics has emerged as an indispensable tool in the education sector, offering insights that can improve student outcomes and inform more equitable policies (Friedler et al., 2019; Kleinberg et al., 2018). However, the widespread adoption of predictive models is hindered by several challenges, including the lack of accessibility, the potential perpetuation of inequalities, and the introduction of bias during various stages of modeling. This paper presents a multifaceted approach to address these challenges, focusing on the development of fair, transparent, and interpretable predictive models for educational applications. The proprietary nature of many predictive models restricts the ability of researchers and practitioners to evaluate, adapt, or optimize these models to align with ethical considerations and accountability standards, undermining the principles of transparency and fairness in high-stakes domains such as education (Haghighat et al., 2024). Education is a complex and dynamic system that requires accounting for the diversity, variability, and uncertainty of data and context. Therefore, predictive models in education must be transparent, interpretable, and fair to ensure accountability, trustworthiness, and ethical design (Haghighat et al., 2024). The challenges in deploying predictive and statistical techniques in practice are manifold, presenting at different steps of modeling, including data cleaning, identifying important attributes associated with success, selecting the correct predictive modeling technique, and calibrating the hyperparameters of the selected model. Each of these steps can introduce additional bias to the system if not appropriately performed (Nezami et al., 2024). Purpose/Objective/Research Question: The objective of this research is to assess unfairness of the predictive models in important stages of data cleaning, modeling, and model selection and develop techniques to more fair, transparent, and interpretable for educational decision-making. Specifically, we aim to: 1. Develop a fair predictive model that integrates fairness measures into the learning process, ensuring equitable treatment across different demographic groups. We introduce a novel approach using multivariate adaptive regression splines (MARS) that embeds fairness constraints within the model's architecture. This method not only secures fairness in the selection of knots but also yields interpretable decision rules, enhancing the model's transparency and adaptability for various stakeholders (Haghighat et al., 2024). 2. Evaluate the influence of imputation techniques on the performance and fairness of predictive models in the context of college student success. We scrutinize the impact of different imputation methods on the equity of predictive outcomes, using a comprehensive national education dataset. Our investigation reveals the critical role of responsible data preprocessing in preventing the introduction of bias, thereby ensuring that predictive models are just and representative of all student subgroups (Nezami et al., 2024). 3. Create an interactive system that supports fairness-aware model selection and hyperparameter optimization. We present FairPilot, an innovative exploratory tool that incorporates Fair and Accurate Bayesian Optimization (FABO). This system empowers practitioners to navigate the complex trade-offs between model accuracy and fairness. By providing a user-friendly interface, FairPilot enables the selection of optimal model configurations that align with a diverse array of fairness definitions, ensuring that machine learning applications in sensitive domains are both precise and just (Di Carlo et al., 2024). Setting: We use a large-scale national education dataset, specifically the Educational Longitudinal Study 2002 (ELS), to analyze and predict college student success. The focus is on assessing and addressing the unfairness and accuracy of machine learning models used for predicting academic outcomes in higher education. The empirical evaluations and comparisons are conducted using categorical and numerical success variables. Population/Participants/Subjects: The study population included students from various 4 year educational institutions (public and private), with a focus on college students for the assessment of predictive modeling outcomes. Intervention/Program/Practice: The intervention involved the development and implementation of fair predictive models, the assessment of imputation techniques, and the creation of an interactive system for fairness-aware model selection and hyperparameter optimization. Research Design: The research design combined quantitative and qualitative methods to evaluate the fairness, transparency, and interpretability of predictive models. Strategies for eliminating sources of bias included the use of fairness metrics and the incorporation of interpretability into the model design and model selection processes. 1. The research design focuses on the development and evaluation of the fairMARS method, which extends the capabilities of MARS to prioritize fairness in knot selection and coefficient estimations. The study conducts fairness-accuracy trade-off analysis using different values of a fairness parameter ([lambda]) within both fairMARS and fair Decision Tree (DT) models. Empirical evaluations are performed on datasets, including the ELS, to compare the performance of fairMARS with existing baselines and to analyze the effects of fairknot and faircoef subroutines on fairness and accuracy. The design includes a practical case study to compare the resulting basis functions and coefficients across models and to observe the balance between fairness and accuracy. 2. The research design involves a prospective evaluation to assess the impact of imputation techniques on the fairness and accuracy of predictive modeling outcomes for college student success. The study uses ELS dataset and applies various imputation strategies, including Multiple Imputation (MI), KNN Imputation, and the removal of rows with missing values (Remove-NA). Fairness metrics such as Statistical Parity (SP), Predictive Equality (PE), Equal Opportunity (EOP), and Equalized Odds (EO) are used to assess unfairness across racial groups. The study compares the performance and fairness of different ML models using various imputation strategies and explores the impact of imputation on unfairness in different perturbation scenarios. 3. The research design involves the development and application of FairPilot, an interactive system for fairness-aware model selection and hyperparameter optimization. The system uses multi-objective Bayesian optimization (MOBO) and grid-search techniques to explore the trade-off between accuracy and fairness in predictive modeling. Experiments are conducted using the ELS dataset to demonstrate the features and functionalities of FairPilot, evaluate the effectiveness of Fair and Accurate Bayesian Optimization (FABO), and explore the trade-off between accuracy and fairness across different hyperparameters. The study aims to support practitioners in choosing fair and accurate models for different scenarios and to justify the integration of FABO in FairPilot compared to state-of-the-art methods.