Internet of Behavior (IoB) and Explainable AI Systems for Influencing IoT Behavior

by   Haya Elayan, et al.

Pandemics and natural disasters over the years have changed the behavior of people, which has had a tremendous impact on all life aspects. With the technologies available in each era, governments, organizations, and companies have used these technologies to track, control, and influence the behavior of individuals for a benefit. Nowadays, the use of the Internet of Things (IoT), cloud computing, and artificial intelligence (AI) have made it easier to track and change the behavior of users through changing IoT behavior. This article introduces and discusses the concept of the Internet of Behavior (IoB) and its integration with Explainable AI (XAI) techniques to provide trusted and evident experience in the process of changing IoT behavior to ultimately improving users' behavior. Therefore, a system based on IoB and XAI has been proposed in a use case scenario of electrical power consumption that aims to influence user consuming behavior to reduce power consumption and cost. The scenario results showed a decrease of 522.2 kW of active power when compared to original consumption over a 200-hours period. It also showed a total power cost saving of 95.04 Euro for the same period. Moreover, decreasing the global active power will reduce the power intensity through the positive correlation.



There are no comments yet.


page 3

page 4

page 7


Toward Organic Computing Approach for Cybernetic Responsive Environment

The developpment of the Internet of Things (IoT) concept revives Respons...

Energy Efficient Service Distribution in Internet of Things

The Internet of Things (IoT) networks are expected to involve myriad of ...

Intelligence Stratum for IoT. Architecture Requirements and Functions

The use of Artificial Intelligence (AI) is becoming increasingly pervasi...

Autonomic Management of Power Consumption with IoT and Fog Computing

We introduce a system for Autonomic Management of Power Consumption in s...

An 8-bit In Resistive Memory Computing Core with Regulated Passive Neuron and Bit Line Weight Mapping

The rapid development of Artificial Intelligence (AI) and Internet of Th...

Constraint Programming Algorithms for Route Planning Exploiting Geometrical Information

Problems affecting the transport of people or goods are plentiful in ind...
This week in AI

Get the week's most popular data science and artificial intelligence research sent straight to your inbox every Saturday.

1 Introduction

Natural disasters and epidemics have affected people’s lives for centuries. Tsunami, Spanish flu, HIV/AIDS, COVID-19 pandemic, and others have caused millions of deaths worldwide and forced people to change their behaviors. Its impact on increasing mortality, causing economic damage, changing the behavior of individuals, causing social and economic disruption, and increasing political pressure and tensions which compels developed countries to analyze them carefully.

While changing behavior appears to be a low-risk effect compared to mortality or economic disaster, it is powerful enough to cause such risks. For example, in a study of the effect of social distancing on suicide rates apart from epidemiological deaths, results showed that a change in the social distancing index of ten units caused a 2.9% increase in the suicide rate [13]

. Moreover, during the COVID-19 pandemic, scientists reported that wearing a mask could prevent approximately 130,000 deaths in the USA alone. Many behavioral aspects have been altered by the COVID-19 pandemic, such as customer interaction with brands, employee work procedures, and business engagement with consumers. All of these and other examples have an economic, technological, and physiological effect. Therefore, tracking people’s behavior becomes crucial to influence it in adverse situations. For example, using machine learning for mask recognition tasks is one way to get individuals to respect regulations and monitor negligence.

By 2023, It is predicted that the activities of 40% of the global population will be tracked digitally to influence the human behavior [7]. It is also projected that the IoT sensors market will reach $22.48 billion and 29.3 billion internet-connected devices will be available by 2023. Therefore, IoT would play a major role in detecting people’s behavior [5].

The Professor of Psychology, Gote Nyman, was the first to announce the concept of the Internet of Behavior (IoB). In 2012, he said that if the human behavioral pattern is assigned to devices (e.g. IoT devices) with specific addresses, there will be an opportunity to benefit from the knowledge gained by analyzing the history of patterns in many businesses, societal, health, political, and many others fields. Behavior is a psychological characteristic that can determine a person’s willingness to cooperate or collaborate. Despite the other characteristics: Cognition, Emotion, Personality, and Inter-Communication, behavior is responsible for the tendency to act and is highly dependent on the other four criteria. Therefore, focusing on the behavior will allow us to know how to influence and treat the person.

Attempting to influence and change user behavior is sensitive, as it may encounter resistance and other psychological factors related to comfort and trust. Emerging technologies such as XAI will help provide the user with the required understanding and trust of any system that uses AI models. XAI aims to use methods and techniques to explain to the user what an AI model does and why, thus giving a better perception of system operations. Therefore, the process of tracking, analyzing, and influencing user behavior will become much simpler.

This paper integrates the IoB concept with the XAI concept to propose trusted and understandable frameworks that work in users’ behavior-changing areas. Accordingly, it proposes an IoB-based XAI-based system implemented in a use case scenario of household electrical power consumption that aims to change the consumer behavior into an eco behavior to reduce power consumption and therefore reduce power waste and cost. This framework integrates IoT, AI, Data Analytics, Behavioral Science, and XAI techniques to achieve user and business benefits. The contributions of this paper can be summarized as follows:

  • To the best of our knowledge, this is the first attempt to investigate the concept of IoB, it’s workflow and benefits, challenges, and current industrial directions.

  • Propose a trusted and comprehensible IoB-XAI-based system that attempts to influence and change IoT behavior to achieve users and businesses benefits.

  • Present a use case scenario of electrical power consumption behavior to influence consumer behavior and raise awareness to control and reduce power consumption process.

  • Discuss and compare the main characteristics of related work in power consumption area.

  • Highlight potential developments and future directions of the proposed system.

2 Internet of Behavior (IoB)

IoB is using IoT devices to collect a massive amount of human behavioral data and turn it into valuable insights to improve user experience and search experience by changing users’ behavior, interests, and preferences.

IoB operates in three areas, IoT devices, User/Search experience, and Behavior. It connects devices to the internet to track user behavior, then analyzes this behavior to improve search and user experience in order to influence their behavior to reach a certain result.

IoB attempts to appropriately understand data and use this understanding to create new products, promote for current products, redesign the value chain, increase profits, or reduce costs from a psychology perspective. Therefore, user behavior in the IoB workflow will be tracked first using connected devices, as shown in Fig. 1. The data generated by IoT devices will be collected and then analyzed using data analytics and machine learning algorithms. The analysis phase will yield useful information that must be properly understood from a behavioral science perspective. Finally, the knowledge gained will be used to develop business strategies and influence the behavior of users, thus achieving a specific goal.

Fig. 1: IoB Workflow

As shown in Fig. 2, the IoB process is descriptive yet proactive, it tracks and analyzes user behavior in order to detect and accordingly influence important psychological variables to achieve the process objective. This usually brings lots of benefits as well as a huge set of challenges.

Fig. 2: IoB Process

2.1 IoB Benefits

IoB is a very new concept in today’s emerging technologies. It’s being used in many applications for several benefits, however, without much knowledge about it in the research community. Therefore, in this sub-section, we summarize IoB benefits:

  • QoE with Increased Profit: Help companies resolve issues in targeting more sales while keeping their customers satisfied at the same time (Win-Win strategy).

  • Tasks Automation: Help replace outdated fashion tactics like time-consuming and unfavorable customer surveys (No More Tedious Tasks).

  • Target Customers: Give the opportunity to know the valuable customers based on their interest and which segment to target and invest more time with.

  • Accuracy: Give the ability to track and study the unobtainable behavior of how customers interact with products and services (Notice the Unnoticed).

  • Real-Time Interaction: Help provide real-time interactions via notifications or alerts to customers about targeted offers, sales, and/or advertisements.

2.2 IoB Challenges

2.2.1 Security

Dealing with sensitive and real-time data always creates security concerns for any service user. In IoB, it is necessary to demonstrate a high level of security and protection to users, as their behavioral data is expected to be prone to attacks. The data sensitivity motivates cybercriminals to access, reveal, collect and benefit from it.

2.2.2 Ethical Use

Behavioral data is a sensitive and personal data type, and its collection, storage, or analysis must be accompanied by transparency and ethical use. The user has the right to be aware of this process as well as to know that his privacy is preserved and protected from misuse. Another aspect of ethical use is not following the principle that the end justifies the means, with companies in most cases aiming for more profit, regardless of the impact on the user of changing his behavior in critical areas such as health. Last, in the ethical challenges, is to ensure the consent of the user’s when his data is being collected and used for any purposes.

2.2.3 Ostrich Effect

The ostrich effect is a phenomenon that occurs when the rational mind believes something is important and the emotional mind expects it to be painful. People understand the power and benefits of using IoB. However, they may be uncomfortable (trust issues) about the tracking process, which results in avoiding or rejecting it. Therefore, when companies implement IoB, it is not enough to work on process understanding by users. They also have to work on removing the discomfort of using this technology.

2.3 XAI for IoB

IoB challenges may create resistance from the users’ side as it tries to influence and change their behavior and track their data. Using AI models in IoB to analyze users’ data allows taking advantage of XAI techniques to make the operations of IoB-based systems more intelligible to the user and thus more trustworthy.

Describing the system process, workflow, impact, and outcomes using user-friendly mathematical and analytical proofs will give the users a better understanding of the system and increase awareness of its impact. Thus, ease the problems of trust and the impact of resistance.

3 Eye on Industry

In this section, we will discuss how the different industries are interacting with IoB technology.

3.1 Healthcare

Nowadays, smartphones allow many software companies to build and promote different applications that care about users’ health where they can easily use smartphone sensors to track bio-metrics and healthy behavior. For instance, the Beddit health app tracks sleep patterns, heart rate, and breathing rate. After analyzing the tracked behavior, the app sends alerts, notifications, and tips to improve the user’s sleep and motivates him/her to achieve daily goals to reach a more positive outcome [3].

During the COVID-19 pandemic, many countries have developed health smartphone apps for citizens to control the spread of the virus. The China Health Code Alipay app, for example, tracks a user’s travel, contact history, and body bio-metrics such as temperature. Then, a colored QR code is generated to identify his health status. Accordingly, some restrictions may apply to the user and affect his behavior, such as allowing travel or quarantine at home or in a central location [9].

3.2 Entertainment

YouTube recommendations are a popular system that utilized users’ behavior, especially in viewing history and clicks, to improve the experience on an app or a website. Using deep neural networks, their algorithm analyzes the collected data and tries to recommend videos and shows that a user might be interested in. In addition, almost all social media applications are using the same strategy in providing feeds based on the viewing patterns, amount of time spent in one account, or interest in subscribing to certain channels.


3.3 Transportation

Even in transportation, IoB can be applied for benefits, whether on drivers or passengers. After a long history of disputes with drivers and a high turnover rate, Uber is trying to resolve these issues and settle disputes in its favor by influencing driver behavior using gamification. Uber has tricks like loss aversion, recognition, and intrinsic motivation to reward drivers or make them fear losing their winnings [8]. Also, Ford is expanding its reach by joining forces with Argo AI to develop autonomous vehicles that adapt and behave differently to road infrastructure designs and driving behaviors. This technology is being tested in many locations such as the streets of Miami, Washington, and others [2].

4 IoB and XAI For Power Consumption Reduction: A Use Case Scenario

This section proposes a use case scenario of a system for a household electrical energy consumption that aims to influence the electricity consumption behavior of a house by automatically controlling the amount of global active power.

The system workflow in Fig. 3 shows that the current will pass through an energy monitor and controller device (EMC) to know how much electricity each IoT device is using. After that, the data will be transmitted from the EMC to a cloud database to be stored and normalized then used by an AI model to predict the global active power that may be consumed in the next hour. The prediction result will be passed to a decision maker and explainer to determine how much energy the appliances should use and then the results will be returned to the EMC control the consumption in the next hour if it is above a certain threshold in order to reduce it.

The aim of this system is to reduce power consumption that in turn will reduce waste and cost of potential used power by changing consumption behavior in a way that the user understands and trusts.

Before discussing the proposed system flow in detail, a set of related proposals in electrical power consumption area will be discussed. Table I shows a summary of each related proposal and compared to the proposed system (last row in the table). Studies [1], [11], and [15]

, used deep learning algorithms: Long short term memory (LSTM), Convolutional neural network (CNN), or both to predict the electrical consumption. However, they focused their contribution on the architecture of the proposed model. Likewise,

[14] and [10] used LSTM and CNN together.

The researchers in [14] have proposed to predict power consumption through proposing a framework consisting of data cleaning and model building steps. Whereas the authors in [10] aimed to accelerate the deployment process of deep learning models using the proposed edge-cloud framework.

On the other hand, in [12] and [4]

applied Deep Reinforcement Learning algorithms (Deep RL) on the same dataset to perform tasks related to process optimization of power consumption. While authors of

[12] focused on the proposed RL model structure, authors of [4] proposed a stochastic game framework that controls the interaction between the consumed power and household.

Fig. 3: System Workflow

The proposed model of this paper presents a comprehensive framework consisting of all technology needed to form a successful IoB/XAI model which will reduce power consumption by using IoB and XAI to influence user behavior. Firstly, the LSTM algorithm will be used to build a power consumption prediction model, and then the power consumption process will be controlled. Meanwhile, the workflow will be explained to the user to ensure his understanding for a smoother behavior change process.

Ref. Algorithm Dataset Goal Proposal / Framework description
Short-term individual households
power consumption for Australian
Predicting the individual household electrical load. Focus on deep learning model architecture
[11] LSTM
Monthly electricity consumption
from a Korean region
Forecast the annual electricity consumption
based on the history of the last three years.
Focus on deep learning model architecture
Electrical meters datasets of non
residential building in Switzerland
Forecasting building load using Resnet to enhance
the ability of learning
Focus on deep learning model architecture
Individual household electric power
consumption In France /
Appliances Energy Prediction
Predicting electricity consumption at various
distribution and transmission systems’ levels.
Framework of two steps:
data cleaning and model building
Household power consumption
by Korea Electric Power Corporation.
Accelerating a deep learning model
deployment after power data stream training
Framework for a power flow edge
cloud system
[12] Deep RL
Electrical submeters recorded by
Pecan Street. Inc
Optimizing home energy management by taking
a series of actions
Focus on deep RL architecture
[4] Deep RL
Electrical submeters recorded by
Pecan Street. Inc
Scheduling energy consumption for
household appliances
Framework for the interaction of power
grid and households using a stochastic
Individual household electric power
consumption In France
Influencing electricity consumption behavior by
automatically controlling the amount of power.
IoB and XAI based framework of four
steps: Data, Database, Data Normalization,
AI model, Decision making, and XAI
TABLE I: Related Work Comparison

4.1 Dataset

An individual household electric power consumption dataset available online has been used to implement the use case. It contains measurements of one house’s electrical power consumption per minute for approximately 4 years between December 2006 and November 2010, collected in a house located in Sceaux, France. Measurements have been divided between kitchen appliances, laundry room appliances, electric water heaters, and all other household appliances. It consists of 2,075,259 samples. The data has been re-sampled to get the total consumption per hour which reduced the dataset size to 34,589. 50% of the dataset has been used as train samples and the remaining 50% as test samples.

4.2 Database and Data Normalization

Electrical power consumption data is acquired from smart energy monitoring sensors connected to the internet. An advanced type of smart monitoring sensor can detect the type of each connected appliance by connecting the monitor to the electricity meter only, without the need for multiple monitoring IoT sensors to be connected to each appliance. This helps centralize the system and reduce its installation cost.

The acquired data is huge and in its raw format. Therefore, it required a proper database to handle it. For example, MongoDB is one of the NoSQL databases that can handle this data as it is document-oriented, which means it can deal with semi-structured data where most IoT data exists in semi-structured or unstructured formats. Moreover, it has high performance in data storage and retrieval, which is necessary for second-by-second data acquisition and transmission. Furthermore, it is highly available, scalable, dynamic, flexible, and ideal for dealing with heterogeneous data. These features are best for increasing the volume and multiple resources data.

Due to the possibility of incomplete, unstructured, and inconsistent data as IoT data is unstructured, the data normalization phase comes to normalize and pre-process the data to get it in a suitable format without noise and outliers. Additionally, this phase handles data loss that may occur due to disconnection or dropouts, so it deals with missing values by adding data averaging or removing these values. Moreover, it scales features using scaling methods like Min-Max-Scaler to map data to the same range.

4.3 Power Consumption Prediction Model

The dataset used in the scenario is time-series-based data. Time series prediction problems are one of the difficult types of predictive modeling problems because they add sequence dependence complexity amidst input features. The Recurrent Neural Network (RNN) is a robust type of deep neural network that can handle such problems. The Long-term memory network (LSTM) is a type of RNN used in time series problems because it can successfully train huge structures with eliminating major problems, such as vanishing gradients.

A sequential deep learning model has been built to predict the hourly power consumption, it allows building a layered model with each layer having weights corresponding to the following layer. Keras library has been used for coding the model structure. The model has been constructed using one LSTM layer and one Dense layer and ran for 20 epochs with Adam optimizer. The model has been trained on a machine with the following specifications: Intel CPU@2.3GHz, 2 cores, RAM 13GB, and Disk 70GB.

4.4 Decision Maker and Explainer

As the EMC will track the appliances’ power consumption per hour and pass the tracked data to a cloud database to be stored, then normalized and used by the AI model to analyze the consumption behavior and predict the power consumption for the next hour, and based on prediction and historical data patterns, the decision maker part in the Decision Maker and Explainer will decide the amount of power that should be controlled used in the next hour. To personalize the user experience, the decision maker will make the decision based on the user’s behavior and usage pattern. Therefore, the decision will be made by comparing the predicted consumption with the average consumption at specific times related to the prediction hour. Take into account influences seasonal and occasional factors.

The explainer part in the Decision Maker and Explainer is responsible for applying the XAI techniques to describe the AI model, its impact, biases, and outcomes to the user. The purpose is to use a set of processes and methods to allow the user to understand and trust the results generated by the model, thereby raising their awareness of the workings of the system and its goal in order to have a smooth behavior change process.

The explainer will attempt to cover the following concerns:

  • Model prediction process.

  • Model success and failure.

  • Model confidence in decision-making, using data annotation and deep learning algorithm annotation.

Data annotation describes statistical information of the dataset such as features distribution and correlation. While the algorithm annotation shows the characteristics of the model and whether it is biased towards certain features or outliers. The results will be sent to the user whenever they are requested using any internet-connected device.

Accordingly, the entire system works to influence the user’s consumption behavior in his favor by saving power from waste, saving the cost of wasted power, and increasing his awareness of better consuming behavior. Therefore, creating ‘’collective wisdom” and more responsible habits that affect not only the individual but also the society as a whole.

4.5 Use Case Results

4.5.1 Prediction Model Loss

The proposed model loss started with a training loss of 0.0125 and a testing loss of 0.0079. Both of them continued to decline, reaching a training loss of 0.0095 and a testing loss of 0.0078. Fig. 4 illustrates the loss performance over epochs.

Fig. 4 shows a snapshot of the model’s perdition results for the global active power and the actual global active power values of 200 samples. As shown, the prediction follows the actual values pattern, while the error rate is low.

[Model train vs. test loss] [Model results]

Fig. 4: Prediction Model Performance

4.5.2 Power Saving

Since the proposed system predicts the global active power for the next hour and controls the amount of used power to a certain threshold, a certain amount of power will be saved. The system will control the power if the prediction is above the average consumption of the same hour of the same month and the same week day for the past four years. The system has been tested on 200 samples from the testing data to know the amount of the saved power. As shown in Fig. 5, the system will issue warnings (red squares) on 27 hours from 200 predicted hours where the consumption is above the threshold. The total saved power amount for the 27 hours is 522.2 kilowatts.

Fig. 5: System Warnings

4.5.3 Cost Saving

As the system is tested on 200 samples and saved 522.2 kilowatts of global active power, the power cost will also be saved. In France, the price of electricity in kilowatt-hours for households is 0.182 euros. Therefore, the system will save €95.04 for a period of 200 hours.

4.5.4 Global Intensity Reduction

The energy intensity indicates the quantity of energy required per unit output or activity to produce a product or service. A high intensity indicates high energy used to provide a service and low intensity indicates lower energy to provide a service. This affects the price or cost of energy conversion as low energy intensity indicates a low cost of energy conversion into the gross domestic product (GDP). In the dataset, the global intensity is positively correlated with the global active power by 99% so that they both move in tandem, which means that if the global active power decreases, the global intensity will also decrease.

To sum up, by influencing the power consumption behavior with IoB, and with the help of XAI the proposed system in the use case scenario managed to reduce 522.2 kilowatts of global active power in the house over a 200-hours period, while helping to reduce €95.04 from wasted power cost. Moreover, the reduction of the global active power positively affected the global intensity positively which will benefit the community as well. All results will help reshape user behavior into an eco behavior.

5 Future Directions

In the future direction, there are various open issues and areas of potential development that need to be investigated and tested:

Incorporating a user feedback subsystem into the proposed system is an important area for improving the learning and prediction of the AI model and the overall system. It is also important to measure the user satisfaction factor. This will help improve the experience, fix emerging issues, and keep the user more engaged.

Another potential development area is building a distributed version of the proposed system. The aim is to compare the distributed version to the current centralized version and test its impact on both technical and user experience aspects. Such as its effect on network efficiency, cost, data storing process, system and IoT failures, user trust, privacy, and user and IoT behavior.

Moreover, sending out instructive notifications to users to teach them about eco behavior is also a considerable approach in product and system design. Their response and use of instructive notices will speed up the process of changing behavior and make it easier, clearer, and more entrenched. It will change their mentality to wasteful consumption.

Furthermore, emphasizing the use of different AI/ML learning techniques is another development area to reduce the complexity of the AI model and increase its accuracy.

Finally, integrating technologies that add more control over security and data privacy is significant, especially as it will address the IoB security challenge. Therefore, the application of different security schemes, techniques, and technologies for different levels and processes of the system from data transmission to its storage and use will definitely add more value to the system.

6 Conclusion

Since Internet-connected devices are increasingly being used; tracking, understanding, and influencing users’ behavior to achieve material and moral profits has become a reality. This paper is the first attempt to investigate the concept of IoB, it’s workflow and benefits, challenges, and current industrial directions. It also proposes a trusted and comprehensible tracking-analyzing-influencing behavior system using IoB and XAI. Moreover, a use case of an electrical power consumption system that works for benefit of individuals, power companies, and society to save power waste and cost. The collected results demonstrated that the system was able to save 522.2 kW of wasted power and €95.04 of its cost. Furthermore, the positive correlation between active power and power intensity will reduce the latter if the former is reduced.