For stock market investors, precise stock forecasts lead to improved decisions and greater profits. This study focuses on advancing stock return predictions through a clustering approach applied to selected stocks. Deep learning (DL) and Machine Learning (ML) models are trained on representative data from each cluster to forecast stock returns within the clusters. We explore the combination of three clustering methods [Dynamic Time Warping (DTW), K-Means, and Density-Based Spatial Clustering of Applications with Noise (DBSCAN)] with four prediction models [Long Short-Term Memory (LSTM), Convolutional Neural Network (CNN), Random Forest (RF), and eXtreme Gradient Boosting (XGBoost)]. The effectiveness of each DL and ML model combined with clustering is tested using the Diebold-Mariano (DM) test, confirming their superiority over methods lacking clustering. We also introduce a two-step process for portfolio optimization using predicted returns. High-predicted return stocks are initially preselected, followed by applying the Mean-Variance with Forecasting (MVF) optimization model to allocate stocks optimally. Using data from the New York Stock Exchange's large-cap stocks (2013 to 2022), results demonstrate the superiority of each clustering-based prediction model combined with MVF. The findings are further validated through daily trading simulations on test data from 2017 to 2022.