Profound subject knowledge is necessary for education to spark creative or formative projects and boost individual’s dynamism. Acquiring a magnificent education is crucial because it permits us to live completely, make wise decisions, and overcome obstacles in life. A learner’s academic journey is continued in early education, which also permits them to become more adept at discriminating between inequity and fairness and equips them for a broader spectrum of activities in higher education. Alongside superficial constraints like threats or financial crises, learners must compose a plethora of delightful and demanding activities in addition to their studies. Therefore, learners can experience moments of cognitive disorientation and progressively acquire cognitive ailments such as stress or depression, to name a couple. Thus, this research suggests a novel Bi-directional Encoder Representations from Transformer (BERT) method for ascertaining learners’ cognitive declines. This research utilizes the BERT-Base architecture, and the hyper-tuning of the hyperparameters is carried out by the innovative Tweaked Shuffled Frog Leaping Algorithm technique. Although there is a high need for language models these days, there are currently no specific language models that can reliably specify a learner’s cognitive deficits according to psychological inqueries. Thus, this work offers a novel strategy known as CogniBERT in response to this. Our approach confounds some of the weaknesses of the BERT methodology by choosing the right scaling factors during training. In terms of training, validation, and testing accuracies, the experimental result surpasses the other approaches, obtaining 99.56%, 98.97%, and 99.29%, respectively, when the suggested method is contrasted with a few state-of-the-art methods.