Comparing Newton Raphson and Stochastic Gradient Descent Methods for Traffic Accident in Malang

The severity of traffic accidents in Malang Regency has shown a significant upward trend over the past five years, necessitating an in-depth analysis to support safety policy development. This study aims to compare the effectiveness of the Newton–Raphson and Stochastic Gradient Descent (SGD) estimation methods in a binary logistic regression model for predicting accident severity. Secondary data were obtained from the Malang Police Department for the period 2020–2024, consisting of 13 predictor variables and a binary dependent variable (severity: minor–severe). The model was estimated using the Newton–Raphson and SGD algorithms, with a convergence criterion of ε = 10⁻⁶ and a learning rate of α = 0,0001. The results indicate that the Newton–Raphson method produced an AIC value of 3.198,6 and a McFadden’s pseudo-R² of 0,019, while the SGD method yielded a lower AIC of 3.201,9 and a pseudo-R² of 0,018. The variables "student occupation" and "provincial area" were found to be statistically significant (p 0.05). In conclusion, the SGD method demonstrates a slight advantage over the Newton–Raphson method in terms of AIC performance.