司法量刑是人民對司法體系信任度的重要影響因素，但目前國內相關的量刑實證研究仍多所缺乏。本研究主張，我國的量刑改革應重視「法學取向之法事實研究」。本文針對2018年至2020年間《槍砲彈藥刀械管制條例》第7條第4項、第8條第1項與第4項、第12條第4項之1，341件確定判決進行量化統計分析，發現量刑普遍集中於法定刑下限甚至更輕，累犯雖常見但影響力低，而「情堪憫恕」、「自首」與「犯罪後之態度」等減輕因子對刑度影響顯著，反應法官量刑偏向寬緩。以機器學習模型所預測的刑期與真實情形平均誤差僅4.42月，顯示此量刑模式具高度可預測性。此外，釋字第775號對累犯規定實務適用影響有限，亦反應對法律議題的理論興趣與實務落差。整體而言，本研究說明法事實研究對強化量刑理論與實務的重要價值。

Sentencing practices significantly influence public trust in the judicial system, yet empirical research on sentencing in Taiwan remains scarce. This study argues that sentencing reform in Taiwan should prioritize“jurisprudential-oriented empirical legal research,”especially through quantitative statistical analysis of big data. Accordingly, this article conducts a quantitative empirical analysis of 1,341 finalized judgments from 2018 to 2020 under Articles 7(4), 8(1) and (4), and 12(4) of the Controlling Guns, Ammunition and Knives Act. The findings reveal that sentences generally cluster around the statutory minimum or are even lower. Although recidivism is common, its impact on sentencing is minimal. Conversely, mitigating factors such as“extenuating circumstances,”“voluntary surrender,”and“post-offense attitude”significantly influence sentencing, reflecting a general judicial tendency toward leniency. The XGBoost model used to predict sentence duration yielded an average error of only 4.42 months compared to actual sentences, indicating high predictive accuracy. Moreover, Judicial Yuan Interpretation No. 775 has had limited practical effect on the application of recidivism provisions, highlighting a gap between theoretical interest in legal topics and actual judicial practice. Overall, this study demonstrates the critical value of empirical legal research in strengthening both sentencing theory and practice.