Recommendation systems are crucial for providing services to the elderly with Alzheimer's disease in IoT-based smart home environments. Therefore, we present a Reminder Care System to help Alzheimer patients live safely and independently in their homes. The recommendation system is formulated based on a contextual bandit (CB) approach to tackle dynamicity in human activity patterns. Correct recommendations aimed at meeting user needs without their feedback is achieved. Our experimental results show the feasibility and effectiveness of RCS in real-world IoT-based smart home applications.