Objectives: Traditional diagnostic methods in Korean Medicine, such as pulse diagnosis and symptom-based pattern identification, rely significantly on subjective clinician judgment, which limits reproducibility. Meanwhile, conditions like atopic dermatitis (AD) and functional dyspepsia (FD) involve complex neurophysiological mechanisms that could benefit from objective, data-driven evaluation. This study aimed to develop machine learning–based biomarkers from resting-state functional MRI (fMRI) in AD patients and pulse waveform clustering in FD patients, with the goals of (1) distinguishing patients from healthy controls, (2) predicting treatment response, and (3) assessing concordance between data-driven and clinician- or questionnaire-based pattern classification. Methods: In the AD study, 41 patients and 40 healthy controls underwent resting-state fMRI. Signals from 38 functional brain regions were analyzed using a long short-term memory (LSTM) model to extract temporal dynamics, with bootstrapping and 4-fold cross-validation to evaluate classification performance for disease identification and acupuncture treatment response. In the FD study, one-minute pulse waveforms from five channels on each wrist of 100 patients were recorded using multi-channel tonometry. After normalization, Time Series K-Means clustering with Dynamic Time Warping (DTW) was applied in both hard and soft clustering modes. Clustering outcomes for Cold Heat and Deficiency–Excess patterns were compared to clinician and questionnaire diagnoses using accuracy, cosine similarity, and Kullback–Leibler divergence. Results: For AD classification, the LSTM model achieved accuracies ranging from 73% to 85% across key regions— including the supplementary motor area, posterior cingulate cortex, temporal pole, precuneus, and dorsolateral prefrontal cortex—exceeding the chance level of 50% with statistical significance (p < 0.05). For acupuncture response prediction, accuracies reached 71% to 83%, with most predictive regions being the lingual–parahippocampal–fusiform gyrus, primary motor and somatosensory cortex, paracentral lobule, and frontal gyrus. In FD patients, pulse waveform clustering showed higher agreement with clinician-assessed patterns than questionnaire based patterns. Pulse waveform clustering achieved a highest accuracy of 0.75 for Cold–Heat classification at the fifth channel of the left wrist and 0.60 for Deficiency–Excess classification at the first channel of the right wrist when compared with clinician diagnoses.