This paper introduces an improved version of the Akaike information criterion (AIC), termed AIC_C, for selecting smoothing parameters in nonparametric regression. The authors compare AIC_C with other methods, including plug-in methods and classical approaches like generalized cross-validation (GCV) and the AIC. AIC_C is shown to be more stable and less prone to undersmoothing compared to these methods. The paper also presents Monte Carlo simulations demonstrating that AIC_C performs well in various scenarios, including when plug-in methods fail. The AIC_C criterion is derived and evaluated for different nonparametric regression estimators, including local polynomial, convolution kernel, and smoothing spline estimators. The results show that AIC_C is competitive with plug-in methods and performs well in situations where such methods are not effective. The paper concludes that AIC_C is a useful and reliable method for selecting smoothing parameters in nonparametric regression.This paper introduces an improved version of the Akaike information criterion (AIC), termed AIC_C, for selecting smoothing parameters in nonparametric regression. The authors compare AIC_C with other methods, including plug-in methods and classical approaches like generalized cross-validation (GCV) and the AIC. AIC_C is shown to be more stable and less prone to undersmoothing compared to these methods. The paper also presents Monte Carlo simulations demonstrating that AIC_C performs well in various scenarios, including when plug-in methods fail. The AIC_C criterion is derived and evaluated for different nonparametric regression estimators, including local polynomial, convolution kernel, and smoothing spline estimators. The results show that AIC_C is competitive with plug-in methods and performs well in situations where such methods are not effective. The paper concludes that AIC_C is a useful and reliable method for selecting smoothing parameters in nonparametric regression.