The distributed computing environment allows to provide the outsourced computing services in addition to web-services for IoT and mobile technologies. An emerging research topic is the QoS and security indicator prediction to achieve a reliable service selection that meets user requirements. Collaborative filtering technique is one of the most widely used model in service selection. It is based on similarity computation between users or services. But the main drawback of this method is the lack of data to compute an effective similarity value. Furthermore, malicious users give false feedback which influences the accuracy of prediction. In this work, we propose a novel similarity evaluation model based on self-organisation map to address the problem of data lack and the robust index computation to detect the untrustworthy users. The proposed approach uses a K-means-based average evaluation to determine the tenderness of the data and an offline build-up model to increase computational efficiency.