Oil well stimulation measures are the main technical means to increase oilfield production. Due to the diverse types of measures and influencing factors, there has been a lack of efficient prediction methods to guide rational production allocation and economic evaluation for a long time. Research has shown that although old oil fields have accumulated a large number of measure samples through long-term development, there are problems such as lack of prominent main control factors, long training cycles, and unsatisfactory prediction results with direct application of conventional machine learning algorithms. This article proposes a method for predicting the effectiveness of measures based on collaborative filtering, which screens cases with similar distances and high similarity in multidimensional factor spaces through collaborative filtering. Using top ranked cases as learning samples instead of the entire sample set and then building prediction models through machine learning algorithms such as KNN and SVM, avoids directly using conflicting rules from the entire set of samples. Deep learning algorithms have high degrees of freedom, weak generalization ability, and significant outlier predictions. After comparative testing, the average prediction accuracy of three independent and composite measures, namely extraction, acid, and perforation, for the development of old oil fields with medium and high permeability through water displacement is over 80%, which is significantly reduced compared to the outlier prediction results of conventional algorithms and has value for field promotion and application.