Combining citation and productivity metrics through harmonic mean enhances researcher ranking accuracy

Addressing the challenge of predicting scientific impact and ranking researchers is a complex yet critical task, drawing significant attention from scholars across diverse fields. This effort plays a key role in improving research productivity, supporting decision-making processes, and advancing methodologies for scientific evaluation. Over time, various metrics such as citation counts, total publications, hybrid methods, the h index, and h-type indicators have been introduced to identify influential researchers. Despite these efforts, no single metric has been universally accepted as the best approach, as different metrics serve varying purposes and contexts. This study presents a novel index developed through comprehensive analysis of a dataset comprising 1060 Neuroscience researchers, including both awardees and non-awardees. The initial phase of the research involved evaluating specific metrics to determine their ability to place awardees among the top 100 researchers, leading to the identification of the five parameters most frequently associated with awardee inclusion. Advanced deep learning techniques were then applied to refine the selection, pinpointing the top five influential parameters and assessing the disjointness in their outputs. To further enhance the findings, seven statistical models were examined for their ability to combine the most disjoint parameter pair while retaining their individual strengths. Selecting the most disjoint pair ensures that the ranking process integrates diverse evaluation criteria rather than relying on redundant or highly correlated parameters. This approach captures a broader spectrum of researcher impact, reducing bias and increasing the robustness of the final ranking index. Among these models, the h2 upper and k indices exhibited the highest disjointness ratio at 0.97. Additionally, the Harmonic Mean approach demonstrated superior performance, achieving an average impact score of 0.76, and excelled at preserving the unique features of the selected parameter pair. Based on these results, a new index was formulated using the Harmonic Mean (HM) of the most disjoint pair. This index showed significantly improved performance compared to existing metrics, offering a robust solution for ranking researchers effectively.