Commonly known as Darija, sentiment analysis in the Moroccan Arabic dialect has special difficulties because of the lack of defined writing norms, grammatical complexity, and small annotated datasets. In this work we provide SENTIMAROC, a novel approach for real-time sentiment analysis combining machine learning methods with a web application. Our approach consists on data collecting from social media platforms, thorough preprocessing to clean and normalize the language, and the deployment of models including Support Vector Machines, Random Forest, and Logistic Regression. Latent Semantic Analysis (LSA) and TF-IDF were among the advanced feature extraction methods used to improve classification accuracy. With an overall accuracy of 94%, the models exceeded benchmarks in Moroccan sentiment analysis on a dataset of 32,101 tagged comments. Real-time sentiment prediction on the accompanying web application offers a simple interface. This work not only shows the success of the suggested strategy but also greatly helps sentiment analysis in underfunded languages like Darija to be advanced.