Leaf Group Formulation Preference Algorithm Based on Historical Replacement Data and Two-Stage Optimisation Model

This article proposes a method for optimizing the recommendation of tobacco leaf formulations based on historical replacement data analysis and a two-stage optimization algorithm. Firstly, a six-dimensional attribute consistency scoring model is constructed using historical data, and efficient data acquisition and preprocessing are carried out with the open-source tool Kettle, which includes data cleaning and structuring. Secondly, by combining replacement behavior trajectory clustering with attribute consistency scoring, the method assesses the substitutability among materials, thereby further optimizing the material recommendation process. Finally, experimental validation indicates that the proposed optimization recommendation algorithm outperforms traditional methods in both accuracy and applicability, effectively aiding material replacement decisions in practical production.