Real Steps or Not: Auto-Walker Detection in Move-to-Earn Applications

TL;DR

This paper introduces an AI method to detect fake walking in Move-to-Earn apps, ensuring rewards are given only for real physical activity.

Contribution

A novel AI-based approach is proposed to distinguish genuine user movement from auto-walker simulations in M2E platforms.

Findings

The model achieved an F1-score of 0.997 on auto-walker datasets.

It attained an F1-score of 1.000 on genuine gait datasets.

The method generalizes well to unseen datasets.

Abstract

In recent times, the emergence of Move-to-Earn (M2E) applications has revolutionized the intersection of digital innovation and physical wellness. Unlike their predecessors in the Play-to-Earn (P2E) domain, M2E apps incentivize physical activity, offering rewards for real-world movement such as walking or running. This shift aligns with a growing global focus on health consciousness that is propelled by the widespread adoption of smartphones and an increased awareness of the benefits of maintaining an active lifestyle. However, the rising popularity of these platforms has also brought about new problematic activities, with some users exploiting additional automated devices to simulate physical activity and claim rewards. In response, we propose an AI-based method aimed at distinguishing genuine user engagement from artificially generated auto-walker activity to ensure the integrity of reward distributions in M2E platforms. To demonstrate the generalizability of our model, we use a total of six open gait datasets and auto-walker datasets of automatic walking devices measured with various smartphones. Under unbiased and transparent evaluation, our model shows its ability to effectively discriminate auto-walker and genuine gait data not only on the seen datasets but also on the unseen datasets; it attained an F1-score of 0.997 on the auto-walker datasets and an F1-score of 1.000 on the genuine datasets.

Figures