Background: Pediatric infections are often non-specific and clinical diagnoses rely much on history taking from family. We applied Natural Language Processing (NLP) to evaluate family reported outcome and to develop a novel syndromic surveillance for influenza-like illness among pediatric patients. Methods and materials: Patients’ vital signs and family reported outcome from chief compliant at emergency triage were analyzed to extract nigh major syndrome groups of influenza-like illness: (1) fever, chills and sweating, (2) headache, cry and irritate, (3) nose, (4) mouth and throat, (5) eye, (6) ear, (7) skin rash, (8) gastrointestine (9) appetite and activity. Neural Network (a machine learning algorithm) was compared with domain expert keywords. The trends of the 9 syndrome groups were compared with clinical diagnoses at discharge among children visiting emergency department at a medical center in Taiwan, from Jan 2017 to Dec 2018. Results: During the two-year period, there were totally 29,233 events of pediatric emergency visits, and 15,414 (53%) were clinical diagnosed influenza-like illness (ICD-9: 460–488). Boys (56%) and young children were more dominant in respiratory infections: infant (16%), toddlers (48%), and preschooler (20%). The Pearson correlation between trend of clinical diagnosed respiratory infections and trend of body temperature ≥38 °C was 0.84. The correlation increased when using syndrome group (1) “fever, chills and sweating”, which achieved a higher correlation at 0.94, and the correlation further increased to 0.95 (p < 0.05) when combining syndrome groups (1) to (4). Conclusion: Patients’ vital signs and family reported outcome at initial encounter can be applied as an effective and timely syndromic surveillance to predict influenza-like illness at emergency department.