<p>QoS-aware service composition suffers from many problems due to the uncertainties of its working environment. Besides, the many candidate services for service composition consume more processing and execution time. This paper presents a four-phases framework for reliable dynamic Web service composition. The preprocessing phase helps to reduce the percentage of missing values of the dataset. Then, clustering phase applies self-organizing maps (SOM) algorithm to cluster the QoS data to reduce data sparsity and cold-start problems of web services. Then, the prediction phase predicts uncertain QoS values of services using an optimized LSTM deep learning model. In the composition module, an improved hybrid optimization algorithm IHHO-SWO is the proposed integrating the Spider Wasp Optimizer (SWO) and is applied to generate a composite service that fits the end user requirements. An exhaustive experimental and statistical study validates the superior performance of the suggested framework. The study reveals that the framework is able to successfully deal with service prediction and composition challenges such as outlier handling, cold start, and data sparsity, lowering these problems considerably.</p>