Novel Dynamic Batch-Sensitive Adam Optimiser for Vehicular Accident Injury Severity Prediction

2026-05-14 · Source: Machine Learning · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Expert, quick

Summary

A novel optimiser, Dynamic Batch-Sensitive Adam (DBS-Adam), has been developed to enhance deep learning model efficiency and convergence, particularly for imbalanced and sequential datasets. DBS-Adam dynamically adjusts the learning rate based on a batch difficulty score, which is calculated from exponential moving averages of gradient norms and batch loss. This mechanism increases updates for challenging batches and reduces them for easier ones, improving training stability and accelerating convergence. The optimiser was integrated into Bi-Directional LSTM networks for vehicular accident injury severity prediction, a task that also utilized SMOTE-ENN resampling and Focal Loss to address class imbalance. Rigorous evaluation against AMSGrad, AdamW, and AdaBound across five random seeds showed DBS-Adam achieved competitive performance, with a statistically significant precision improvement (p=0.020), 95.22% test accuracy, 96.11% precision, 95.28% recall, 95.39% F1-score, and a test loss of 0.0086.

Key takeaway

For Machine Learning Engineers developing models on imbalanced or sequential datasets, adopting DBS-Adam can significantly improve training stability and convergence speed. Your models could achieve higher precision and overall performance, as demonstrated by its 96.11% precision in accident severity prediction. Consider integrating DBS-Adam, especially when working with Bi-Directional LSTMs and techniques like SMOTE-ENN and Focal Loss, to enhance real-time classification capabilities and support critical applications.

Key insights

DBS-Adam dynamically scales learning rates based on batch difficulty, improving stability and convergence for imbalanced sequential data.

Principles

• Optimiser choice significantly impacts model efficiency.
• Dynamic learning rates improve training stability.
• Batch difficulty scores can guide learning rate adjustments.

Method

DBS-Adam dynamically scales the learning rate using a batch difficulty score derived from exponential moving averages of gradient norms and batch loss, increasing updates for difficult batches and reducing them for easier ones.

In practice

• Integrate DBS-Adam with Bi-Directional LSTMs.
• Apply SMOTE-ENN and Focal Loss for class imbalance.
• Evaluate optimisers across multiple random seeds.

Topics

• Dynamic Batch-Sensitive Adam
• Vehicular Accident Prediction
• Deep Learning Optimizers
• Bi-Directional LSTMs
• Imbalanced Data Learning

Best for: AI Engineer, AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• Novel Dynamic Batch-Sensitive Adam Optimiser for Vehicular Accident Injury Severity Prediction — DBS-Adam dynamically scales learning rates based on batch difficulty, improving deep learning performance on imbalanced, sequential data.
• Scaling Learning-based AEB with Massive Unlabeled Data — Stabilized meta-feedback semi-supervised learning effectively scales AEB using massive unlabeled data, significantly improving safety metrics in real-world deployment.
• Improving Crash Frequency Prediction from Simulated Traffic Conflicts Using Machine Learning Based Microsimulation — ML-based microsimulation improves traffic conflict realism for crash frequency prediction.
• Singularity-aware Optimization via Randomized Geometric Probing: Towards Stable Non-smooth Optimization — S-Adam stabilizes non-smooth deep learning optimization by dynamically adjusting step sizes based on local geometric instability, preventing gradient chattering.
• Asymmetric-Loss-Guided Hybrid CNN-BiLSTM-Attention Model for Industrial RUL Prediction with Interpretable Failure Heatmaps — A hybrid CNN-BiLSTM-Attention model with asymmetric loss improves turbofan RUL prediction and interpretability.
• Optimizers in Deep Learning: From Gradient Descent to Adam — Optimizers guide neural network training by efficiently updating parameters to minimize loss, evolving from basic GD to adaptive Adam.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine Learning.