Global growth in participation in higher education has helped to increase diversity of students, and traditionally underrepresented minorities on gender, income and math skills have expanded their presence in engineering education. Nevertheless, late dropout has increased and the number of engineering graduates remains low in western world. The analysis of educational trajectories using process mining techniques can help to explain the relationship between a sequence of academic results and late dropout. This case study seeks to answer how gender, income and entry math skills may explain differences on educational trajectories of engineering students in high-failure rate courses that lead to late dropout. Academic records for 794 engineering students at Universidad Austral de Chile that belongs to cohorts 2007 to 2009, were extracted and analyzed using process mining discovery techniques. Models of educational trajectories on high-failure rate courses were created and then analyzed using the Investment Model as a reference framework. Findings reveal the following: late dropout is related to the number of consecutive semesters that a student maintain pending failed courses; low-income students and those with low entry math skills tend to be more persistent, even if they have unsatisfactory trajectories; female students tend to be more risk-averse when they have unsatisfactory results. Understanding the educational trajectories of students who end in late dropout can help managers and policy makers to improve the curriculum, entry conditions and programs that support disadvantaged students.