Sleep quality of an individual has a pivotal role in maintaining his physical and mental well-being. Measurement of sleep involves segmenting sleep periods into different sleep stages, Non-rapid Eye Movement, Rapid Eye Movement, and Awake. Polysomnography involving electro-graphic measurements of the brain, eye movements and chin muscles together with cardiac and respiratory activity is the golden standard for clinical assessment of sleep. But the clinical setup required is not practicable at home. Our study proposes a Long Short-Term Memory model for classifying sleep stages by combining demographic (age, gender) data and physiological data collected using wearable devices. The dataset used in our study consists of time series data collected from 56 dementia patients. The model achieved approximately 80% accuracy. This work has potential application in remote healthcare monitoring with resource constraints, which makes it an ideal setup to be used in rural India. Further, the model can be enhanced with anomaly detection and personalized recommendation using reinforcement learning techniques aimed at improving the overall sleep quality.