Heart failure (HF) is commonly a chronic condition associated with frequent hospital admissions. Early knowledge about a possible deterioration of this condition would enable early treatment for the prevention of adverse events and related hospital admissions. In this paper we present a computational method for predictive information extraction from daily physiological signals, which can be obtained by a telemonitoring system with wearable sensors. It is based on wavelet analysis of temporal signal patterns. Experiments with data from patients enrolled in a telemonitoring protocol show that the proposed method is capable of predicting HF hospitalization events one day before they happen, even in the case of low compliance to the protocol. These results indicate a promising perspective towards a monitoring system that would provide improved life quality for HF patients.