Predicting process behavior in terms of the next activity to be executed and\or its timestamp can be crucial, e.g., to avoid impeding compliance violations or performance problems. Basically, two prediction techniques are conceivable, i.e., global and local techniques. Global techniques consider all process behavior at once, but might suffer from noise. Local techniques consider a certain subset of the behavior, but might loose the "big picture". A combination of both techniques is promising to balance out each others drawbacks, but exists so far only in an implicit and unsystematic way. We propose LoGo as a systematic combined approach based on a novel global technique and an extended local one. LoGo is evaluated based on real life execution logs from multiple domains, outperforming nine comparison approaches. Overall, LoGo results in explainable prediction models and high prediction quality.

Top