Human activity recognition has progressed notably, focusing on data collection and activity detection techniques. Expanding this domain, our research explores the next activity prediction (NAP). We propose a system leveraging LSTM models to forecast users upcoming activities using historical sensor and activity data. The system also generates personalized recommendations for the associated predicted activity like sensor type, place, and location in a smart home, ranking them based on historical patterns and users’ behavior. This adaptive weighting mechanism refines recommendations with new data inputs, ensuring highly relevant suggestions. We have used a LSTM model giving us accuracy of 74% for User A and 88% for User B for next activity prediction. Detailed classification reports evaluate the recommendations, demonstrating the model’s robust performance and effectiveness in providing tailored, user-centric recommendations.