<p>Ageing and disability have relevance for a fall in the capability to perform activities of everyday routine and a lack of physical exercise that disturbs physical and mental health. Remote monitoring of older and disabled people residing in smart homes is complicated. Human activity recognition (HAR) is an effective area of research for classifying applications and human movement in many places. HAR data are collected in wearable gadgets that consist of several types of sensors or with mobile sensor assistance. As a result, a disabled or ageing individual can rely on a HAR system that monitors activity patterns and interventions in the event of serious events or behavioural changes. Recently, several studies have applied deep learning (DL) techniques in HAR, demonstrating excellent performance in classifying HAR. This study proposes an Intelligent Hyper-parameter Tuning Using Pelican Optimisation Algorithm for the Real-Time Activity Recognition (IHTPOA-RTAR) method for disabled individuals. The aim is to develop an effective and robust method for HAR using advanced DL models. Initially, the min-max normalization is used in the data pre-processing phase to transform and normalize raw input data to improve performance. Moreover, the tunicate swarm algorithm (TSA) model is utilized for feature selection. Furthermore, a hybrid of a convolutional neural network autoencoder with an attention mechanism (CNN-AE-AM) model is employed for the HAR classification process. Finally, the parameter fine-tuning process is performed by implementing the pelican optimizer algorithm (POA) model to improve the classification performance of the CNN-AE-AM classifier. A wide-ranging experimentation of the IHTPOA-RTAR methodology is performed under the WISDM dataset. The comparison analysis of the IHTPOA-RTAR methodology portrayed a superior accuracy value of 99.02% over existing models.</p>