The rapid progress in clinical data management systems and artificialintelligence approaches enable the era of personalized medicine. Intensive careunits (ICUs) are the ideal clinical research environment for such developmentbecause they collect many clinical data and are highly computerizedenvironments. We designed a retrospective clinical study on a prospective ICUdatabase using clinical natural language to help in the early diagnosis ofheart failure in critically ill children. The methodology consisted ofempirical experiments of a learning algorithm to learn the hiddeninterpretation and presentation of the French clinical note data. This studyincluded 1386 patients' clinical notes with 5444 single lines of notes. Therewere 1941 positive cases (36 % of total) and 3503 negative cases classified bytwo independent physicians using a standardized approach. The multilayerperceptron neural network outperforms other discriminative and generativeclassifiers. Consequently, the proposed framework yields an overallclassification performance with 89 % accuracy, 88 % recall, and 89 % precision.Furthermore, a generative autoencoder learning algorithm was proposed toleverage the sparsity reduction that achieved 91% accuracy, 91% recall, and 91%precision. This study successfully applied learning representation and machinelearning algorithms to detect heart failure from clinical natural language in asingle French institution. Further work is needed to use the same methodologyin other institutions and other languages.