Where do they go next? Causal inference-based prediction and visual analysis of graduates' first destination

Predicting graduates' first destinations in advance is crucial for targeted career planning and strategic curriculum development. The relationship between first destinations and academic performance exhibits typical high-dimensional networks and varying degrees of correlation characteristics. To reveal this complex relationship, we developed a prediction and visual analysis method based on causal inference for predicting graduates' first destinations. First, we collaborated with university administrators to conceive the First Destination and Academic Performance (FDAP) dataset, which aims to define the available attributes related to academic performance and first destination. Second, we propose a key feature selection method based on the causal inference and random forest (CIRF), which both extracts key features in FDAP for training models and reveals causal relationships in the FDAP dataset. Third, we propose a novel prediction model, CIRF-MLP, which combines the CIRF method with a multilayer perceptron neural network that can predict students' first destination based on their current academic performance. When benchmarked against four other baseline prediction models on the real FDAP dataset, our model showcased exceptional performance, and the ablation experiment results demonstrate the necessity of each model component. Fourth, we developed CausalCareerVis, a visual analysis system built on the CIRF-MLP model, to analyze the causality and correlation between graduates' First Destination and Academic Performance, and to predict their first destinations based on academic performance. Three case studies highlighted the effectiveness and practicality of our work.