In this thesis, several Machine Learning (ML) methods and hybrid algorithms that were developed are applied for the first time to predict microbial activity during composting. Modeling biological activities for an inadequately understood domain is a difficult task. This thesis evaluates, compares, and analyzes the improved results of the models created and the methods used. The results indicate with statistical significance that hybridizing an eager learner with a lazy one improves learning performance in this domain. Lazy-eager hybrids can form complex, irregular hypotheses. They are suitable because the expressive power of the eager learner is significantly enhanced due to their ability to represent the target function by combining several complex locally approximated hypotheses. The study also showed hybrid rule-based methods and trees to be good performers.