The motivation is starkly quantitative. The authors note that residents of Malaysia’s Klang Valley collectively waste more than 250 million hours each year stuck in traffic, an economic drag equivalent to between 1.1 and 2.2 percent of national GDP. Congestion also inflates vehicular emissions, placing Malaysia among the highest CO2 emitters in ASEAN. Traditional remedies such as building more roads have repeatedly run into the Downs-Thomson paradox, in which added capacity simply induces additional demand. The alternative is proactive management through Intelligent Transportation Systems, which depend fundamentally on accurate short-term traffic forecasting to enable dynamic routing, variable toll pricing, and efficient resource allocation.

Deep learning has become the dominant approach to that forecasting problem, and for good reason. Long Short-Term Memory networks, designed to preserve information across long sequences without the vanishing gradient problems that plague ordinary recurrent networks, excel at capturing the daily and weekly rhythms of traffic flow. But the Malaysian team identified a subtle weakness: standalone recurrent architectures struggle to model the instantaneous cross-correlations between heterogeneous vehicle classes, such as the synchronized surges of passenger cars and heavy commercial vehicles within a single time window. Their solution was to prepend a one-dimensional convolutional layer that acts as a trainable spatial filter, extracting localized inter-class dependencies before the temporal memory cells process the sequence.

The hybrid CNN-LSTM engine was validated against an unusually rich dataset: five years of multivariate toll records from PLUS Malaysia Berhad covering 18 plazas in the Klang Valley, totaling 73,891 aggregated observations across five vehicle classes. The data was normalized with min-max scaling and transformed into a supervised learning problem using a 24-hour sliding window, with a strict 70:15:15 chronological split to prevent data leakage. The convolutional head compresses the raw multivariate input into latent spatial features, the LSTM encodes their temporal evolution, and a final ensemble layer built on XGBoost or Random Forest refines the decision boundary using the latent vector produced by the neural backbone.

The benchmarking results were decisive. The CNN-LSTM consistently outperformed standalone LSTM, Bidirectional LSTM, and GRU baselines, achieving the lowest Root Mean Squared Error and Mean Absolute Percentage Error across the heavy commercial and specialty vehicle classes. The contrast was most dramatic for Class 5 machinery, where the CNN-LSTM posted an RMSE of roughly 744 while the GRU’s error ballooned to nearly 2,071, with a MAPE above 52 percent. Standalone recurrent models, the authors argue, simply cannot cope with the sporadic, non-uniform distributions of heavy machinery and buses, whereas the convolutional front end maps localized spikes before they reach the memory cells, keeping forecasts within a tighter band.

Accuracy alone, however, was never the point. The study’s central innovation is its treatment of explainability as a triangulated, multi-perspective exercise rather than a single post-hoc calculation. Most transportation research relies on SHAP, a game-theoretic method that assigns each feature a marginal contribution to a prediction, but the authors argue that using SHAP in isolation leaves the system vulnerable to that method’s inherent biases. Their framework instead cross-validates three mathematically independent explainers: SHAP, Integrated Gradients, which integrates gradients along a path from a baseline input to the actual input, and LIME, which fits a local linear surrogate around each prediction. Only when these methods agree can the attribution be trusted.

The triangulation paid off in a striking way. Both SHAP and Integrated Gradients overwhelmingly identified Class 1 passenger vehicles and total volume as the dominant drivers of the model’s internal state, with normalized importance scores between 0.98 and 1.00, while LIME alone would have suggested a far more distributed picture. That consensus then served as the bridge to the symbolic layer. The team discretized the continuous attribution scores into a binary formal context, using median-based thresholds to avoid both over-saturation and sparsity, and applied Formal Concept Analysis to extract a concept lattice, a hierarchical structure in which each node represents a maximal set of traffic observations sharing a common set of attributes.

The lattice yields deterministic implication rules with remarkable confidence levels. One rule states that when passenger cars, large trucks, buses, and total volume simultaneously exceed their operational thresholds at a given toll plaza, a spike in small lorries follows with absolute certainty. Another provides the reciprocal logic: surges in heavy commercial traffic inherently signal corresponding surges in passenger car volume. Because these symbolic rules align with the numerical consensus from SHAP and Integrated Gradients, the authors can demonstrate what they call neuro-symbolic alignment: the neural network has genuinely learned to base its predictions on the same foundational variables that govern the human-readable logic, rather than exploiting spurious correlations that no one can audit.

The practical implications extend beyond academic curiosity. The full pipeline, from convolutional filtering through ensemble regression to multi-perspective attribution extraction, achieves an average inference latency of 14.2 milliseconds per 24-hour look-back window on an NVIDIA RTX 4090, comfortably within the 15-minute to 1-hour cycles typical of dynamic traffic management. For highway operators, that means forecasts can be paired with explicit rules for proactive intervention, such as anticipating gridlock before it manifests, and that algorithmic actions in automated tolling or lane allocation can always be backed by a symbolic reason. The authors also argue the architecture is network-agnostic: because it relies on multivariate sequences rather than rigid graph topologies, adapting it to a new corridor requires only retraining the feature extractor on local sensor streams.

The study, published as open access in Applied Intelligence, positions itself as a benchmark for deploying explainable AI in high-stakes transportation infrastructure, and it arrives amid a broader shift in the field toward neuro-symbolic systems that pair connectionist learning with formal logic. The authors acknowledge limitations and chart future work in three directions: real-time integration into live traffic control systems, extension to multimodal networks encompassing bus rapid transit and rail flows, and the development of entropy-driven adaptive thresholding so that the concept lattices can continuously evolve with seasonal traffic regimes. For now, the Malaysian dataset demonstrates the core claim: a forecasting system can be simultaneously neural in its handling of noisy real-world data and symbolic in the logic it offers to the humans who must act on it, closing the trust gap that has long separated deep learning from operational deployment.

Subject of Research: Neuro-symbolic explainable AI framework combining CNN-LSTM deep learning and Formal Concept Analysis for interpretable traffic volume forecasting

Article Title: Multi-Perspective explainability in traffic prediction: A neuro-symbolic approach using CNN-LSTMs and concept lattices

Article References: Mohamad Yusof, S. S., Ab. Ghani, H., Md Saleh, N. I., & Khamis, N. (2026). Multi-Perspective explainability in traffic prediction: A neuro-symbolic approach using CNN-LSTMs and concept lattices. Applied Intelligence, 56(16), Article 505. https://doi.org/10.1007/s10489-026-07512-7

Image Credits: AI Generated

DOI: 10.1007/s10489-026-07512-7

Keywords: traffic forecasting, explainable AI, CNN-LSTM, Formal Concept Analysis, neuro-symbolic AI, intelligent transportation systems, SHAP, LIME, Integrated Gradients, deep learning, Malaysia, time-series prediction

The framework at the heart of this work is formal concept analysis, a mathematical theory introduced by Rudolf Wille in 1982 that has since become a cornerstone of knowledge engineering. In its classical form, formal concept analysis takes a formal context, essentially a table stating which objects possess which attributes, and derives from it a lattice of concepts. Each concept pairs a set of objects, called its extent, with the set of attributes they all share, called its intent. The resulting lattice is not just a list; it is a hierarchy, revealing how general and specific concepts nest inside one another. Biologists have used it to classify species, engineers to organize software libraries, and data miners to surface association rules in transaction data.

But the classical theory assumes the table is complete: every cell is a definite yes or no. The moment question marks appear, the elegant machinery stalls. A patient may or may not exhibit a symptom, a customer may or may not have bought a product, and the raw data simply does not say. Researchers have long responded by considering completions, hypothetical fully specified tables consistent with what is known. Among all possible completions, two stand out: the least completion, which fills every unknown with the most conservative answer possible, and the greatest completion, which fills them with the most generous one. Concepts in these two boundary tables bracket the truth, but computing them, and then computing their concept lattices, can be expensive, and the relationship between what they contain and what the incomplete table itself can reveal has remained murky.

That is precisely the gap the Xi’an team set out to close. Their starting point is a family of structures called partially-known formal concepts, introduced in earlier work to capture two intuitive modes of uncertainty: objects that jointly certainly possess a set of attributes, and objects that jointly possibly possess them. These ideas echo the possibility-theoretic reading of formal concept analysis developed by Didier Dubois, Florence Dupin de Saint-Cyr, and Henri Prade, and the interval-set perspective of Ying-Yu Yao, which treats an uncertain set as a pair of lower and upper bounds. Partially-known formal concepts come in three specific flavors, known as SE-ISI, ISE-SI, and ISE-ISI concepts, each defined through different combinations of upper and lower derivation operators that probe the incomplete context from different directions. Together with the classical formal concepts found in the least and greatest completions, they form four types of concepts, each illuminating a different facet of the same patchy data.

The first major contribution of the new paper is a thorough map of the connections among these four types. Because every one of them is ultimately defined by the same upper and lower derivation operators, the authors suspected, and then proved, that deep structural relationships must bind them. They show precisely how a partially-known concept relates to concepts in the least and greatest completions, establishing conditions under which one can be recovered from the other. This is more than an exercise in mathematical housekeeping. It means that knowledge extracted from a completed version of the data, which may be easier to compute in some settings, can be translated back into statements about the partially-known concepts of the raw incomplete table, and vice versa. The incomplete context, in other words, is not a impoverished cousin of a complete one; it carries enough structure to reconstruct much of what the completions would tell you, if you know how to read it.

The second contribution turns this theory into construction methods. The concept lattice of a partially-known structure, the hierarchical arrangement of all its concepts, can be built from the corresponding formal concept lattices of the completions, rather than from scratch. The authors also examine the relationships among the three kinds of partially-known concepts from two complementary viewpoints: the level of single concepts, and the level of entire collections of extents and intents. On both levels they demonstrate explicit methods for mutual conversion, showing how an SE-ISI concept can be transformed into an ISE-SI concept, and how the families of extents and intents of one type relate to those of another. These conversion procedures give practitioners a kind of Rosetta Stone: once any one of the four concept structures has been computed, the others become accessible through systematic translation rather than independent, redundant computation.

The practical payoff arrives in the third contribution: algorithms. Tian and colleagues designed algorithms that acquire partially-known formal concepts directly from already-computed formal concepts, leveraging the theoretical connections to avoid re-deriving everything from the incomplete table. They then stress-tested these algorithms experimentally, and crucially, they did so under different data missing mechanisms, the statistical regimes that statisticians distinguish as missing completely at random, missing at random, and more structured forms of absence. This matters because real datasets do not lose entries uniformly; missingness often correlates with the very attributes that matter most. An algorithm that is stable only under idealized random deletion would be of limited use in practice. The experiments, whose underlying datasets the authors have made available in a public repository, demonstrate that the proposed methods are both feasible and stable across these regimes, with a particularly striking result: when the incomplete contexts contain relatively few concepts, the algorithms exhibit superior performance.

That last finding may sound like a limitation, but it points to a sweet spot that is common in applied settings. Many real tables, especially those built from expert judgment or curated ontologies, are small in object and attribute counts even when riddled with unknowns. For such data, the new approach offers a way to squeeze every defensible drop of knowledge out of what is known, without pretending to know what is not. The work also connects to a broader research program on three-way concept analysis and concept-cognitive learning, in which the same Xi’an group and collaborators worldwide have been developing tools for reasoning under partial information, from attribute reduction in three-way concept lattices to dynamic updating of concepts as new data arrives. The new paper can be read as a unifying chapter in that program, tying together threads that had previously been studied in isolation.

For the wider field of artificial intelligence, the significance lies in the growing recognition that uncertainty is not noise to be cleaned away but structure to be modeled. Machine learning systems increasingly must operate on incomplete knowledge graphs, partially observed relational data, and sparse feature matrices, and formal concept analysis offers a symbolic, interpretable complement to statistical and neural approaches. By proving that the four concept types of an incomplete context form a coherent, inter-translatable system, and by delivering algorithms that exploit this coherence, the study gives knowledge engineers a principled toolkit for a problem that will only grow as data-hungry systems meet an imperfect world. The mathematics of missing entries, it turns out, has its own hidden lattice, and we are only beginning to climb it.

Subject of Research: Connections among formal and partially-known formal concepts in incomplete formal contexts

Article Title: A further understanding for four types of concepts in incomplete contexts based on their connections

Article References: Tian, X., Ren, R., Wei, L., & She, Y. (2026). A further understanding for four types of concepts in incomplete contexts based on their connections. Applied Intelligence, 56(15), Article 460. https://doi.org/10.1007/s10489-026-07311-0

Image Credits: AI Generated

DOI: 10.1007/s10489-026-07311-0

Keywords: formal concept analysis, incomplete formal contexts, partially-known formal concepts, concept lattices, three-way concept analysis, missing data, knowledge acquisition, derivation operators, concept-cognitive learning, data mining, applied intelligence, uncertainty modeling