Machine Learning Techniques in Discrete Choice Modeling

Discrete choice modelling is central to understanding decision-making in fields such as transport planning, marketing, energy consumption and health behaviour. Historically founded on theory-driven frameworks like multinomial and mixed logit models, the discipline has increasingly adopted machine learning to improve predictive performance, accommodate large-scale data and reveal complex non-linear patterns. Methods such as random forests, gradient boosting machines and deep neural networks permit flexible specification of interactions and heterogeneity without exhaustive manual tuning. At the same time, issues of interpretability and class imbalance have spurred the integration of explainable techniques—among them Shapley-based attributions, entity embeddings and layer-wise relevance propagation—to translate algorithmic output into behavioural insights. This fusion of econometric robustness and data-driven adaptability enables practitioners to deploy models on diverse and unstructured datasets, automate feature engineering and streamline calibration. The result is a more scalable and transparent approach to forecasting choice behaviour, with immediate applications in urban mobility management, personalised marketing, demand forecasting for renewable energy and policy evaluation.

Research from Nature Portfolio

No recent Nature Portfolio content available.

Research from all publishers

Recent work has addressed the challenge of class imbalance in travel mode choice by systematically combining six over-sampling and under-sampling schemes with a suite of machine learning classifiers. Results on a metropolitan passenger mode dataset demonstrated substantial gains in minority‐class F1 scores without compromising overall accuracy, underscoring the value of balanced resampling for reliable predictions in highly skewed choice contexts.

A study employing a random forest on census‐block-group data explored how urban built-environment attributes shape commuting mode shares. By coupling the model with SHAP (SHapley Additive exPlanations) values, researchers provided local and global assessments of feature importance for bus, rail, walking and driving choices. This approach resolved the opacity of ensemble methods and delivered policy-relevant insights on density, diversity and destination accessibility.

A recent discussion paper has consolidated the theoretical and practical intersections between traditional choice modelling and modern machine learning. By clarifying conceptual parallels and reviewing case studies—from support vector machines to neural networks—it identified opportunities to adopt automated model selection, integrate text and image inputs, and leverage cross-validation practices. The authors also proposed a research agenda to evaluate when and how data-driven methods can most effectively augment classical utility-based models.

Machine Learning Techniques in Discrete Choice Modeling publication trend

The graph below shows the total number of articles in machine learning techniques in discrete choice modeling across all publications each year (not limited to Nature Index journals).

Technical terms

Discrete choice modelling: A framework for predicting decisions among a finite set of alternatives based on utility maximisation principles.

Over-sampling and under-sampling: Techniques for rebalancing training data by increasing minority-class instances or reducing majority-class instances to mitigate class imbalance.

Random forest: An ensemble learning method that constructs multiple decision trees and aggregates their outputs to improve predictive accuracy and control overfitting.

SHAP values: A game-theoretic approach to explain individual predictions by attributing the contribution of each feature to the model’s output.

Entity embedding: A representation learning technique in which categorical variables are mapped to continuous vectors within a neural network to capture semantic relationships.

References

1. Travel Mode Choice Prediction Using Imbalanced Machine Learning. IEEE Transactions on Intelligent Transportation Systems (2023).
2. Non-Linear Associations Between the Urban Built Environment and Commuting Modal Split: A Random Forest Approach and SHAP Evaluation. IEEE Access (2023).
3. Choice modelling in the age of machine learning - Discussion paper. Journal of Choice Modelling (2022).