A deep learning approach for hiking travel time estimation based on personal walking ability

At the University of Electro-Communications, a research team led by Mizuho Asako, Yasuyuki Tahara, Akihiko Ohsuga, and Yuichi Sei has developed a new deep learning model called “HikingTTE” that significantly improves hiking travel time estimation. Hiking is popular worldwide, but accidents still occur when hikers underestimate the time needed to reach their destination.

The work is published in the journal Cybernetics and Information Technologies.

This model could help reduce mountain accidents and improve hiker safety by providing more accurate travel time predictions. Previous hiking travel time estimation methods often use the relationship between slope (uphill or downhill) and walking speed. However, these methods do not fully take into account individual walking ability or how fatigue builds up over long distances.

HikingTTE addresses these issues by combining a modified Lorentz function-based slope-speed function with a deep learning framework that includes LSTM (Long Short-Term Memory) and attention modules. LSTM is well suited for handling time-series data, while the attention mechanism highlights important parts of the GPS data for more accurate predictions.

A key strength of HikingTTE is its ability to learn a hiker’s walking ability from only part of the GPS data recorded during the trip. By analyzing the performance on the first part of the route, the model creates a slope-speed function for that person and then applies it to estimate the remaining travel time.

Additionally, by using LSTM and an attention-based mechanism, HikingTTE accounts for changes in terrain and the effects of fatigue, leading to more reliable estimates than existing models.

In experiments, HikingTTE outperformed conventional hiking travel time estimation techniques, reducing the Mean Absolute Percentage Error (MAPE) by 12.95 percentage points. It also outperformed other deep learning models originally designed for transportation tasks by 0.97 percentage points. The research team believes that these results could set a new standard for hiking travel time estimation.

In the future, the team plans to include each hiker’s past logs to further personalize the predictions. By helping hikers plan and adjust their pace more effectively, this innovation is expected to prevent delays, minimize risks, and ultimately save lives on the trail. The model could also be integrated into hiking apps or navigation tools, providing practical and reliable guidance.

Provided by
The University of Electro-Communications