Automobile seat comfort prediction: statistical model vs. artificial neural network

Area of Science:

• Automotive Engineering
• Human Factors Engineering
• Computational Modeling

Background:

• Current automobile seat comfort development relies on expensive, outdated trial-and-error methods.
• Process improvements are sought through empirical and prediction models.
• Subjective comfort is a key factor in automotive seating design.

Purpose of the Study:

• To compare the efficacy of stepwise linear regression and artificial neural network models in predicting automobile seat comfort.
• To evaluate which modeling approach better relates objective measures to subjective comfort perceptions.

Main Methods:

• Collected data included seat-interface pressure, anthropometrics, demographics, and perceived seat appearance.
• Developed two predictive models: stepwise linear regression and artificial neural network.
• Related collected data to an overall comfort index derived from a validated 10-item survey.

Main Results:

• Both models adequately predicted subjective comfort perceptions.
• The artificial neural network model demonstrated superior performance with higher R-squared values (0.832 vs. 0.713).
• The neural network model also exhibited lower average error values (1.192 vs. 1.779) compared to linear regression.

Conclusions:

• Artificial neural networks offer a more accurate and reliable approach for predicting automotive seat comfort compared to traditional linear regression.
• Implementing advanced modeling techniques can significantly enhance the efficiency and reduce the cost of automotive seat development.
• These findings support the integration of data-driven prediction models into the seat comfort engineering process.