Fatigue-induced cognitive impairment during prolonged cognitive activity is still a serious concern because it seriously impairs decision-making in fields like surgery, aviation, and transportation that demand extreme precision and attentiveness. The effectiveness of deep neural networks in identifying fatigue from electroencephalography (EEG) signals has been shown in recent developments. Nonetheless, traditional architectures, especially those that use large pooling layers, frequently fail to capture long-range dependencies and retain the complex spatial-temporal information present in EEGderived representations. This study proposes a novel residual network (ResNet) approach to predict decision-making quality using EEG signals under fatigued and non-fatigued conditions. We used an established EEG dataset of twelve healthy participants. In order to process EEG data, the chosen EEG signals are converted into time-frequency spectral representations while maintaining both temporal and spectral features using a Continuous Wavelet Transform (CWT). The resulting spectrograms are concatenated and subsequently passed into the proposed model for classification. However, we present a contrastive convolutional neural network model that was trained on CWT-converted EEG data and is based on Custom ResNet18. The fine-tuned model classifies decisionmaking quality as “high” or “low” with improved generalization and robust performance, achieving an accuracy of 97.03% and 77% for the effective combined subject and cross-subject and 98.87% for each subject’s average accuracy. Our results highlight the effectiveness of the proposed residual network-based framework in accurately predicting decisionmaking quality under prolonged cognitive activity using EEG signals.