General Performance Score for classification problems

Abstract Several performance metrics are currently available to evaluate the of Machine Learning (ML) models in classification problems. ML usually assessed using a single measure because it facilitates comparison between several models. However, there is no silver bullet since each metric emphasizes different aspect classification. Thus, choice depends on particular requirements and characteristics problem. An additional problem arises multi-class problems, most well-known only directly applicable binary In this paper, we propose General Performance Score (GPS) , methodological approach build for The basic idea behind GPS combine set individual metrics, penalising low values any them. users can that relevant based their preferences obtaining conservative combination. Different -based compared with alternatives problems real simulated datasets. built proposed method improve stability explainability usual metrics. Finally, brings benefits both new research lines practical usage, where tailored considered.