The stock market is inherently volatile, making it very hard for investors to create an efficient portfolio. Novel machine-learning (ML) techniques are beneficial in predicting high-performing future stocks and making portfolio selections. This research can change the scenario for investors by saving their time and offering more reliable predictions on what stocks would rock in actuality to take high risks, maximizing returns. Machine-learning regression models like Random Forest, Extreme Gradient Boosting (XGBoost), K-Nearest Neighbors (KNN), and Artificial Neural Networks (ANN) are used to estimate stock values for the next time. Stocks were chosen for this study’s first phase based on their better potential and availability of rewards. The mean VaR portfolio optimization methodology is used to pick a portfolio in the second stage. The performance of the proposed approach is tested on monthly datasets for 29 software companies under the Bombay Stock Exchange Information Technology (BSE-IT) index, India. The mean VaR model combined with RF predictions performs best, as shown in the results section, reaching an accuracy of 0.895, significantly outperforming traditional models (12% Sharpe improvement). Integrating other approaches can improve accuracy and provide a more robust basis for making portfolio selections—ideal in times of high volatility to give an edge on the investment process.