Almost all businesses today use composite web services based on the set of web services working concurrently to attain a goal. Therefore, continuous availability of critical services in SOA is important as it is used by safety critical and business systems. This paper proposes an early predictive and recovery mechanism based on fault-tolerance for early prediction and recovery of scheduled outages (SO). The mechanism allows an efficient outage planning. Once an SO is predicted, the mechanism allows recovery mechanism to be applied according to the service criticality (SC). LMFTSO is comprised of other two models: scheduled outage learning model (SOLM) and fault tolerance learning model (FTLM). These two proposed models are used for learning SO in the context of SBS and FT mechanisms respectively. An explanation-based machine learning (EBL) is used. The proposed model is implemented using PROLOG. A case study of a SOA-based e-commerce has been taken for validation.