Complex Event Recognition (CER) technology, as an important branch of Complex Event Processing (CEP), has driven the development of various related techniques through its progressive research advancements. A new direction in this field is to predict when a pattern might occur before the CER engine detects its actual occurrence, enabling proactive responses to potential events, which is referred to as Complex Event Forecasting (CEF). However, existing studies on CEF have primarily focused on sequential patterns, with limited attention given to effectively describing and handling complex nested patterns. To enhance CEF’s capacity to predict nested patterns with enriched semantics, we propose a method to simplify nested patterns utilizing several flattening rules and employ Symbolic Finite Automaton (SFA) to encode different operators. We also delineate the characteristics of patterns that are conducive to accurate prediction. Additionally, we present two methods for deriving predictions using a probabilistic model, balancing the trade-offs between time and memory costs. Finally, we conduct a comparative analysis of the temporal and spatial efficiency of our optimizations and assess the prediction quality using two models, demonstrating the advantages of our approach.