Hybrid model for driver drowsiness prediction using neuro-fuzzy and random forest

Abstract

In this research, a new hybrid model designed was introduced to identify driver drowsiness by integrating an adaptive neuro-fuzzy inference system (ANFIS) with the random forest (RF) method. Driver drowsiness is one of the major risk factors for road accidents, and timely, accurate identification of it is vital for improving traffic safety. The main emphasis of our work is on the combination of ANFIS with RF. ANFIS uses fuzzy logic to represent intricate systems with uncertainties and non-linear relationships, on the other hand, RF offers powerful classification by its ensemble learning technique, performing well in high-dimensional feature spaces. This combination capitalizes on each model's strengths: ANFIS for its dynamic responding to uncertain data and RF for its high classification accuracy, robustness, and resistance to overfitting. The results from the different experiments confirmed that the hybrid ANFIS-RF outperformed its components considerably. It was able to return an enhancement in accuracy by 2.70% compared to ANFIS, and 0.16% compared to RF, while at the same time, dropping the mean squared error (MSE) by 33.40% compared to ANFIS and 2.85% compared to RF. Knowing that RF is already better than other traditional machine learning methods in classification, these results emphasize its major role in the effectiveness of the hybrid model.