Recent years have seen a sharp increase in the frequency of dementia and related cognitive problems, which poses serious difficulties for caregivers and medical professionals. A typical sign of dementia is agitation, which frequently results in a higher load for caregivers and a worse quality of life for both patients and caregivers. Hence, the goal of this work is to create prediction models for agitation levels in dementia patients using Technology Integrated Health Management (TIHM) dataset. The TIHM dataset consists of continuous monitoring data gathered from wearables, smart home sensors, and other Internet of Things (IoT) devices placed in dementia patients’ living place. This extensive dataset offers insightful information about routines, habits, and surrounding conditions. Several deep learning methods are employed in this study to examine temporal sequences of sensor data and identify patterns suggestive of agitation. These methods include Recurrent Neural Network (RNN), Long-Short Term Memory (LSTMs), LSTMs with attention and pre-trained models. The main goal is to create reliable and accurate prediction models that can anticipate agitation levels in real time. LSTM with attention model achieved an accuracy of 98% and a recall of 100% by outperforming the results of the existing literature. The agitation level forecasting models can revolutionize dementia care through early interventions, personalized treatment plans, and advanced support for caregivers. The ultimate objective is to provide caregivers and healthcare professionals with the tools and insights needed to enhance the quality of life for dementia patients.