This master project investigates the binary classification of backchannels and filled pauses using acous- tic and prosodic features to deepen the understanding of linguistic interactions. Backchannels and filled pauses are key elements in natural speech, indicating listener engagement and speaker hesitation. Accurate classification of these elements is essential for advanced speech processing applications. Two decision tree methods, Random Forest and Conditional Inference Trees, are employed to compare their performance in detecting backchannels and filled pauses. The dataset for this analysis is sourced from the GRASS corpus, a speech database for Austrian German. An analysis of feature importance is conducted to determine which acoustic and prosodic features are most influential in the classification process. The results show that combining these features significantly improves the accuracy of distinguishing between backchannels and filled pauses. This research enhances speech analysis by evaluating classification methods and identifying key features, contributing to the development of more advanced speech processing systems for applications such as automatic speech recognition (ASR).