University of Tehran Journal of Sciences, Islamic Republic of Iran 1016-1104 36 1 2025 01 01 Hybrid Prediction Models for Suicide Mortality Levels in Iranian Provinces: Spatial Econometrics vs. Random Forests 39 49 105959 10.22059/jsciences.2025.397759.1007936 EN Mohadeseh Farzammehr 1 Judiciary Research Institute, Tehran, Islamic Republic of Iran Geoffrey J McLachlan 2 Department of Mathematics, University of Queensland, Brisbane, Australia Journal Article 2025 07 01 With the growing utilization of advanced machine-learning techniques, such as random forests, understanding the significance of spatial factors within these models is increasingly imperative. This study proposes a novel approach to develop spatially explicit classification random forest models by integrating spatially lagged variables, mirroring various spatial panel data econometric specifications. We assess the comparative performance of these models against traditional spatial and non-spatial regression methods to predict suicide mortality rates across 31 provinces in Iran, utilizing data from 2011 to 2021. Results reveal that the spatial random forest model, incorporating spatial lag parameters, achieves a remarkable accuracy of 89.19% in predicting suicide mortality levels, surpassing traditional spatial econometric models (46.51%) and non-spatial random forest models (27.03%). While highlighting the effectiveness of spatial random forest models with spatial lag parameters, this study also recognizes the continued relevance of traditional spatial econometric models in predicting suicide mortality rates. These findings offer valuable insights into the interplay between spatial considerations and predictive modeling, providing essential guidance for researchers in selecting appropriate models for spatial data analysis. https://jsciences.ut.ac.ir/article_105959_946777c125cd78dc9a816ee82fbd1987.pdf