Development of a service parts recommendation system using clustering and classification of machine learning

Abstract

After receiving a service request for malfunctioning gas boilers, service engineers visit the site and provide the appropriate service. However, the same failure phenomena can continually recur when the engineer has a lack of experience or when the cause of the failure is unclear. In response to these situations, this study proposes a machine learning (ML)-based service part recommendation system that can predict the analytic result of a problematic part that has been collected in advance using field service report data registered by service engineers, then recommend the optimal service parts based on the prediction result. First, this method starts with a clustering stage, where engineers are divided into groups according to their skill level by using the K-means clustering algorithm. In the classification stage, this system predicts the intensive analysis result of the problematic part collected in advance, which is generally time consuming. Further, cross validation is performed based on the training data, and the random forest (RF) classifier that shows the best performance is selected. Subsequently, the optimal levels for hyperparameters are derived to increase the performance of the evaluation indices. The optimal conditions for the classification decision are also presented to evaluate various evaluation indices in balance and ultimately increase the recall performance. Finally, in the recommendation stage, the set of parts with the highest service values is recommended for engineers based on data from a group of skilled engineers to achieve a quality enhancement through increasing the life span of the parts and decreasing the cost.