Failures during web service execution may depend on a wide variety of causes. One of those is loss of Quality of Service (QoS). Failures during web service execution impose heavy costs on services-oriented architecture (SOA). In this paper, we seek to achieve a self-healing architecture to reduce failures in web services. We believe that failure prediction prevents the occurrence of failures and enhances the performance of SOA. The proposed architecture consists of three agents: Monitoring, Diagnosis and Repair. Monitoring agent measures quality parameters in communication level and predicts future values of quality parameter by Time Series Forecasting (TSF) with the help of Neural Network (NN). Diagnosis agent analyzes current and future QoS parameters values for diagnose web service failures. Based on its algorithm, the Diagnosis agent detects failures and faults in web services executions. Repair agent manages repair actions by using Selection agent.