Comparison of different hyperparameter optimization methods on driving behavior recognition
DOI: 10.29007/41f5
archive: archived pipeline: cataloged verified
Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)
Summary
This study addresses the challenge of optimizing machine learning models for driving behavior recognition, a critical component for Advanced Driving Assistance Systems (ADAS). Specifically, the authors investigate whether hyperparameter optimization improves the performance of an existing Artificial Neural Network (ANN)-based state machine model designed to recognize lane-changing behaviors. The research compares two optimization techniques—Bayesian optimization and Genetic Algorithms (GA)—against the original, unoptimized model to determine which approach yields the most accurate recognition of lane changes to the left, lane changes to the right, and lane keeping. The methodology utilizes real human driving data collected from seven drivers using a SCAN eRT M driving simulator in a highway environment. Each driver performed 40-minute training sessions and 10-minute testing sessions. The input variables for the models consisted solely of environmental factors, including time-to-collision metrics with surrounding vehicles and operational data such as steering angle, pedal positions, and indicator status. The core model employs three distinct ANN models within a state machine topology to estimate transitions between three states: lane change right, lane keeping, and lane change left. Hyperparameters optimized included the number of hidden layer neurons, activation function, learning rate, and number of epochs. Bayesian optimization used an expected-improvement-per-second-plus acquisition function over 30 iterations, while GA used a population of 25 over 100 generations. Following hyperparameter selection, model parameters (weights and biases) were optimized using the Non-dominated Sorting Genetic Algorithm-II (NSGA-II) to minimize false alarm rates and maximize detection rates. The results indicate that Bayesian optimization generally outperformed GA in terms of recognition accuracy and detection rates across most metrics, likely due to its ability to leverage past evaluations to focus the search space. Bayesian optimization also required less computational time (under 400 seconds) compared to GA (over 470 seconds). However, the most significant finding is that the original model, which used fixed hyperparameters (10 hidden neurons, tansig activation, 100–200 epochs, and a 0.01 learning rate), outperformed both optimized models in the majority of evaluation metrics, including overall accuracy and false alarm rates. While the optimized models achieved moderate to high performance, they did not surpass the baseline model’s efficiency in most scenarios, suggesting that the specific search regions explored by the optimization techniques may not have been optimal for this particular model structure. The significance of this work lies in its empirical comparison of hyperparameter tuning methods within the context of driving behavior prediction. It highlights that while Bayesian optimization is a more efficient and generally superior tuning method compared to Genetic Algorithms for this application, automated hyperparameter optimization does not guarantee improved performance over carefully selected baseline configurations. The findings suggest that for ANN-based state machine models in ADAS, manual parameter selection or alternative optimization strategies may be necessary to achieve optimal recognition performance, as standard optimization techniques may fail to identify the global optimum for complex behavioral models.
Key finding
The original model without hyperparameter optimization outperformed models optimized using Bayesian optimization or genetic algorithms in most recognition metrics.
Methodology
simulator
Sample size: 7
Provenance
The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed. Discovered via author_sweep_intake on 2026-05-28.
| Stage | Outcome | Tool | Model | Prompt | Attempts | Completed |
|---|---|---|---|---|---|---|
| discover | success | author_sweep | — | — | 2 | 2026-05-28 |
| archive | success | unpaywall | — | — | 2 | 2026-06-04 |
| extract | success | cached | — | — | 3 | 2026-06-10 |
| clean | success | clean | — | — | 1 | 2026-06-04 |
| chunk | success | chunk | — | — | 1 | 2026-06-04 |
| embed | success | embed | Qwen/Qwen3-Embedding-8B | — | 1 | 2026-06-04 |
| enrich | success | — | — | — | 1 | 2026-05-28 |
| promote | success | — | — | — | 1 | 2026-06-04 |
| summarize | success | llm | qwen3.6-27b-prismaquant | summ-v5 | 2 | 2026-06-10 |
| tag | success | vector_similarity | — | — | 15 | 2026-06-11 |
| verify | success | — | — | — | 2 | 2026-06-10 |
Summary generated by qwen3.6-27b-prismaquant on 2026-06-10; verification: verified.
Topics
Ranked by relevance to this paper. Hover a topic for its definition.
Information type
What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).
- Theoretical Contribution: computational model