Optimal Pricing of Internet of Things: A Machine Learning Approach

Mohammad Abu Alsheikh, Dinh Thai Hoang, Dusit Niyato, Derek Leong, Ping Wang, Zhu Han

This paper, published in IEEE Transactions on Mobile Computing (and available on arXiv), provides a rigorous economic framework for the "Data-to-ML-to-Service" pipeline that defines the modern Internet of Things (IoT) landscape.

TL;DR

The authors propose a mathematical model to determine the optimal pricing for machine learning-based IoT services. By defining service quality as a function of training data size, they derive closed-form solutions for standalone and bundled pricing and use cooperative game theory (Shapley values) to ensure fair profit-sharing between partners.

What Problem Does This Paper Address?

While we often talk about the technical side of IoT—sensors, latency, and model accuracy—the economics of these services are often overlooked. An IoT service provider sits in the middle of a three-tier market:

  1. Data Vendors: Sell raw sensor data.
  2. Service Providers: Buy data to train ML models and sell "intelligence" (e.g., fraud detection, OCR).
  3. Customers: Pay a subscription fee based on the perceived value and quality of the service.

The core challenge is that service quality is not static. In ML, quality (accuracy) depends on the volume of training data purchased. If a provider buys too little data, the service is poor and customers won't pay; if they buy too much, the data costs eat the profit. This paper provides the "missing link" by mathematically connecting ML performance metrics to market pricing strategies.

Methods

The authors introduce several novel components to solve this optimization puzzle:

  • The Quality Function ($q$): They model service quality using an exponential function: $q(n) = \alpha_1 - \alpha_2 \exp(-\alpha_3 n)$, where $n$ is the data size. This captures the "diminishing returns" law of machine learning—the first 1,000 samples help a lot, but the millionth sample helps very little.
  • Non-Linear Optimization: They formulate profit-maximization problems for both standalone services and bundled packages. Crucially, they provide closed-form solutions, meaning providers can calculate optimal prices and data requirements instantly without needing expensive iterative simulations.
  • Cooperative Game Theory: When two providers bundle their services (like a "Smart Home" package combining security and energy management), they must decide how to split the extra profit. The authors use the Core solution to ensure the partnership is stable and the Shapley Value to ensure the profit split is mathematically "fair" based on each provider's marginal contribution.

Datasets

The authors validated their models using three diverse real-world datasets to represent different IoT tasks:

  1. 20 Newsgroups: Used for text categorization (predicting user context).
  2. Sentiment140: Used for sentiment analysis (wearable devices sensing mood).
  3. MNIST: Used for Optical Character Recognition (OCR) (e.g., a smart fridge reading expiration dates).
  4. Google Surveys: The authors conducted their own empirical survey of 100 users to model the "willingness-to-pay" ($\theta$) distribution, finding it follows a uniform distribution in these contexts.

Key Insights and Ideas

  • The "Aha" Moment on Bundling: The paper proves that bundling is almost always a "win-win-win." The total profit of the coalition is higher than the sum of individual profits, providers can offer a discounted rate to attract more customers, and customers get more value for less money.
  • Data Cost Sensitivity: As the cost of raw data ($c$) increases, providers don't just raise prices; they actually buy less data, which lowers service quality, and they are then forced to lower subscription fees to maintain their customer base.
  • Fairness via Game Theory: The use of Shapley values prevents "free-riding" in partnerships. If Service A contributes more to the bundle's attractiveness than Service B, the math ensures Service A receives a larger share of the pie.

Key Results

  • Model Fit: The proposed exponential quality function showed an excellent fit across different ML algorithms, including Deep Learning, Support Vector Machines (SVM), and Naive Bayes.
  • Profit Gains: Numerical results showed that a bundled service (e.g., Service 1 + Service 2) generated a total profit ($F_K \approx 19.67$) significantly higher than the sum of their standalone profits ($8.31 + 9.58 = 17.89$).
  • Optimal Thresholds: The study identified specific "data saturation points" (e.g., $n^* = 18.68$ units for text categorization) beyond which buying more data becomes economically irrational, even if it slightly improves accuracy.

Limitations and Future Work

  • Monopoly Assumption: The current model primarily looks at a monopolist provider or a cooperative coalition. It does not fully account for "price wars" or aggressive competition between rival providers.
  • Uniform Willingness-to-Pay: While the authors found a uniform distribution for $\theta$ in their survey, luxury IoT services or niche industrial applications might follow different distributions (like Gaussian or Power Law), which would change the optimization results.
  • Future Directions: The authors suggest extending this to "substitute" services (where one service replaces another) and exploring dynamic pricing where data costs fluctuate in real-time.

Summary updated

Abstract

Internet of things (IoT) produces massive data from devices embedded with sensors. The IoT data allows creating profitable services using machine learning. However, previous research does not address the problem of optimal pricing and bundling of machine learning-based IoT services. In this paper, we define the data value and service quality from a machine learning perspective. We present an IoT market model which consists of data vendors selling data to service providers, and service providers offering IoT services to customers. Then, we introduce optimal pricing schemes for the standalone and bundled selling of IoT services. In standalone service sales, the service provider optimizes the size of bought data and service subscription fee to maximize its profit. For service bundles, the subscription fee and data sizes of the grouped IoT services are optimized to maximize the total profit of cooperative service providers. We show that bundling IoT services maximizes the profit of service providers compared to the standalone selling. For profit sharing of bundled services, we apply the concepts of core and Shapley solutions from cooperative game theory as efficient and fair allocations of payoffs among the cooperative service providers in the bundling coalition.
BibTeX
@article{alsheikh2020optimal,
  author = {Mohammad Abu Alsheikh and Dinh Thai Hoang and Dusit Niyato and Derek Leong and Ping Wang and Zhu Han},
  title = {{Optimal Pricing of Internet of Things: A Machine Learning Approach}},
  journal = {IEEE J. Sel. Areas Commun.},
  year = {2020},
  url = {https://openreview.net/forum?id=XhJaV8TsK4},
}

Loading related content...

Figure