M.S. Applied Data Science - Capstone Chronicles 2025

17

reliable early warning predictor can be done by using public school-level data and does not have to be student-data that is private. Due to this, the approach allows for more accessibility from districts of all sizes while also being in compliance with FERPA. There are numerous studies revolving around student level data that this study adds to it at a school level and is helpful in identifying the schools that are experiencing challenges with regards to the system. Finally, the findings from this analysis revealed that using data that is open and not sensitive can be useful in decision making that is more equitable. It will also provide help with early interventions for all of California’s diverse schools and give insight into where to allocate resources. Overall, this study provided insight on the usefulness of school level predictions which are more attainable and practical. With the publicly available data, we were able to combine relevant indicators. The Early Warning System gives districts and schools a tool they can use to confirm when there is support needed and be able to intervene early enough before the graduation rates decrease. 6.3 Recommend Next Steps / Future Studies This EWS has different opportunities in which it can be worked on for future studies. First, gathering and using climate indicators for the schools that are up to date can improve future models. This represents how safe a school feels for students and their sense of belonging, as well as the support they have with staff and other students. With newer and complete CalSCHLS data, the study will be able to expand to better capture the patterns in behavioral and engagement. Another recommendation is to have more information on the support and services dimensions, such as number of counselors that are available, what access do schools have in academic intervention, or how available mental health services are for students. Another suggestion is to use data from several years

6.1 Limitations There were a few limitations that were encountered in this study. One limitation was not being able to use the climate data as it was outdated (2017–2019), reported only at the county level, and it was missing for seven counties. The school climate indicators provide information on the school environment a student is in, this has to do with their feelings of safety, connectedness, and inclusion. These are all important factors that can measure engagement, but due to the limitation it was not useful, and it did not improve the performance of the model. This resulted in dropping the climate variables from the final model, which still performed well and had a more reliable set of predictors. Another limitation was the lack of variation that school level data gave, it did not provide information on more specific grouping of students. Lastly, the data used was all from California, which does not have the same reporting system as other states. There would need to be adjustments made for those states based on their reporting structure and the data that is available. 6.2 Conclusion This study developed a school-level EWS that is capable of predicting low graduation outcomes using only publicly available data that is at the school level from the CDE. The strongest predictors that came from this study were: chronic absenteeism, unexcused absences, FRPM eligibility, still-enrolled rates, and A-G completion. The variables are ones that are common in research that describe how attendance, socioeconomic obstacles, and preparation in academics have an association with graduation outcomes. There was another variable that emerged which was Support and Services. This gave insight into how teacher experience had a moderate effect on graduation rates in a broader sense. The performance of these models had PR-AUC scores between 0.75-0.79. This signifies that a

206