Background Length of hospital stay (LOS) is an important measure of patient flow efficiency through healthcare systems. This study aimed to evaluate the use of complex machine learning, a type of artificial intelligence, to predict LOS and extended LOS after general surgery. Methods A cohort study was conducted using a prospectively maintained database including general surgery admissions at two hospitals in South Australia between January 2017 and March 2023. The primary outcome was LOS, and secondary outcome was long LOS (LLOS), defined as LOS >28 days. Data were cleaned and preprocessed, before feature engineering and modelling within machine learning analyses. Results After data cleaning and preprocessing, 16,938 general surgery admissions remained. Within final machine learning modelling for LOS prediction, performance with respect to R2 value ranged from 0.31 to 0.57 across the eight evaluated models. The most important features included having major surgical complications, complication grade, complexity of surgery, admission type, and general surgery subspecialty. For the prediction of LLOS, final accuracies for the models ranged from 0.94 to 0.98. Across the models, the most important features included surgical complication grade, major complications, and visit frequency. Conclusions This detailed study demonstrated that machine learning algorithms were able to predict LOS and LLOS after a wide range of general surgery. Several machine learning models were evaluated. Moderate performance was observed for the prediction of LOS, and excellent performance was observed for the prediction of LLOS. Surgical complication presence and severity were the most important features within the machine learning modelling.