The Bohai P3 oilfield is a loose sandstone oilfield with complex clay mineral compositions and high content. Long-term water injection has led to the dislodgement and migration of clay mineral particles, causing blockages within the reservoir and near the wellbore area, which not only decreases the oil well production but also brings difficulties to the prediction of oil well output. This paper establishes an oil well production prediction method based on the KM-LLE-XGBoost model, which is highly applicable to the oil well production prediction in the P3 oilfield. Firstly, based on geological static factors, this paper applies the KM clustering algorithm (K-Medoids) to finely divide the reservoir into 3 categories; then, combining 17 influencing factors such as geological static, production dynamic, fluid properties, well network, and oil reservoir parameters, a Spearman rank evaluation model is established to optimize and select the main controlling factors affecting oil well production; finally, the Locally Linear Embedding (LLE) nonlinear dimensionality reduction method is applied to further process the main controlling factors, serving as the input parameters for the prediction model, while also optimizing the hyperparameters of the XGBoost ensemble model, establishing the final KM-LLE-XGBoost oil well production prediction model. After optimization, the prediction accuracy of the KMAfter optimization, the prediction accuracy of the KMGBoost model has significantly improved the prediction accuracy of the KM-LLE-XGBoost model has significantly improved, with the degree of fit between predicted and actual values increasing from 83.21% before optimization to 98.42% after optimization. The model was applied to guide oil increase measures for a total of 68 well times in P3 oil field in 2023, among which there were 64 well times with an oil increase greater than 10 m3/d, reaching a standard compliance ratio of 94.1%, with a single well peak daily oil increase of 28 m3/d, and a water cut decrease of 2.3%.Based on the actual situation of the P3 oilfield reservoir, this article applies machine learning algorithms to finely classify the reservoir and establishes oil well production prediction models for each type of reservoir. It also identifies the most critical factors affecting oil well production, providing technical support for the subsequent stable oil control water work of the P3 oilfield. At the same time, it offers reference and guidance for the oil well production prediction and measure direction of similar loose sandstone oil reservoirs offshore.