Emotion deficit has been found in individuals having Alzheimer’s disease (AD). Electroencephalography (EEG) is effective in emotion recognition. However, wet EEG for emotion recognition can limit its applicability. To address these challenges, this paper explores the use of wearable dry EEG technology, which offers a more usable alternative. The study proposes a novel hybrid network that combines an Echo State Network (ESN) with a Gated Recurrent Unit (GRU) to perform an emotion recognition task. This approach categorises emotions into positive, negative, and neutral. Using a dataset with 2132 samples, the proposed network ‘Echo-GRU’ combining ESN and GRU demonstrates high performance, achieving an accuracy of 95.58%, surpassing state-of-the-art models. The results of this model provide a valuable index for supporting the screening and early diagnosis of AD. By accurately identifying emotional deficits, this approach offers a promising biomarker for healthcare professionals to support the early detection of AD.