A Framework for Prediction and Storage of Battery

Life in IoT Devices using DNN and Blockchain

1st Siva Rama Krishnan Somayaji 2nd Mamoun Alazab 3rd Manoj MK
School of Information Technology College of Engineering, IT and Environment School of Information Technology
Vellore Institute of Technology Charles Darwin University Vellore Institute of Technology
Vellore, India 0909, Australia Vellore, India
[email protected] [email protected] [email protected]

4th Antonio Bucchiarone 5th Chiranji Lal Chowdhary 6th Thippa Reddy Gadekallu
Fondazione Bruno Kessler (FBK) School of Information Technology School of Information Technology
arXiv:2011.01473v1 [cs.CR] 3 Nov 2020

Trento Vellore Institute of Technology Vellore Institute of Technology

Italy Vellore, India Vellore, India
[email protected] [email protected] [email protected]

Abstract—As digitization increases, the need to automate vari- the battery drains out. Once the battery drains out, it needs to
ous entities becomes crucial for development. The data generated be replaced. During this time span, critical data which may
by the IoT devices need to be processed accurately and in a affect the overall analysis and the decisions making process
secure manner. The basis for the success of such a scenario
requires blockchain as a means of unalterable data storage to may be lost and this result in the reduction of profits for the
improve the overall security and trust in the system. By providing organization. The second major hinderance is security of the
trust in an automated system, with real-time data updates to sensitive data. Since the data is transmitted via the internet
all stakeholders, an improved form of implementation takes the to the cloud, there is a chance for an attack in between the
stage and can help reduce the stress of adaptability to complete transmission and also in the storage area. There is a chance of
automated systems. This research focuses on a use case with
respect to the real time Internet of Things (IoT) network which huge threat to the IoT Network as well as the Recommender
is deployed at the beach of Chicago Park District. This real time Systems [4].
data which is collected from various sensors is then used to design As per a recent forecast by Gartner’s, the IoT networks [5]
a predictive model using Deep Neural Networks for estimating and their deployment in the real world are expected to have a
the battery life of IoT sensors that is deployed at the beach. This growth of around 32% by 2021 which results to be around 25.1
proposed model could help the government to plan for placing
orders of replaceable batteries before time so that there can be an billion IoT units installed. This growth must be accompanied
uninterrupted service. Since this data is sensitive and requires to by continued research on the security and scalability issues of
be secured, the predicted battery life value is stored in blockchain Iot networks. The blockchain or Distributed Ledger technology
which would be a tamper-proof record of the data. can be a viable solution to the scalability and security concerns
Index Terms—Blockchain, Internet of Things, Deep Neural of Iot devices [4], [6]–[10]. Blockchain works on the concept
Networks, Battery LIfe.
of shared digital ledger that is continually recompiled to all
the users of the network for every transaction made. Once the
I. I NTRODUCTION transactions are carried out, these transactions are validated
The recent era is moving towards technology-based solu- and then recorded into the digital ledger which cannot be
tions for real-world problems. The intervention of technology modified or deleted. The right to view, amend or transact
in day to day activities is critically high. One of the trending newer information into the ledger is a benefit given to certain
Information and Communication technology is Internet of individuals of the network in the community.
Things (IoT) which is the widely used in the recent decade. Motivation for this work:
IOT is a term that connects people worldwide with places, • The battery life depletion is a major problem in real
products, things, etc, [1]. This also provides a wide opportunity time deployed sensor networks. It can be the major
for value creation and capture. The IoT networks which are contributing factor to why analytic predictions made with
designed are deployed in the real-world environment and the data could not be met
critical data are transmitted to the cloud [3] where the data is • Mission critical data could be lost during the timespan of
stored. Though the world is towards this powerful and enabling the battery replacement.The dependence of batteries are
technology, the deployment of such IoT networks is limited a major inhibiting factor for the depletion of growth rate
and it is not practiced in large numbers in the real environment. in IoT sensor deployment.
The reason for this hesitation in deployment is twofold. The • It can also be seen that the labor cost for monitoring and
first being battery usage of the sensor and the rate at which replacing the batteries overtake the benefits provided by
 the deployed sensors themselves. architecture to verify the edge device by anyone anywhere
• By having a third party take control over the actual task without knowledge of manufactures.
of battery replacement, the trust on the entity could be the O. Novo et. al. [16] proposed a proof-of-concept based
only assurance for their actions. A decentralized ledger on blockchain decentralized system for large distributed IoT
system could provide an immutable copy of the predicted systems to provide scalable access management. Smart contact
battery life thereby combating the possibilities of human contains set operations such as revoke permission, Query
error or scam Manager, Query Permission, remove and add manager. N.
The main contributions of this work are: Fotiou et. al. [17] designed an Ethereum based blockchain
architecture to control a group of IoT devices. This system
• A prediction model is proposed using Deep Neural
uses custom blockchain tokens to build efficient and robust
Networks for predicting the battery life of the sensors
access control methods. X. Liang et. al. [18] proposed public
deployed at six locations in the beach water of Chicago
blockchain architecture along with cloud server to secure
Park District.
drone communication, the transmission of data, preserving
• Securing the sensitive predicted battery life values by
data integrity and cloud auditing. The drone chain contains
storing them in the blockchain to attain a tamper proof
the registration of drone, data transmission, blockchain receipt,
record of this data.
data validation in cloud and decision making. This system
The rest of the paper is organized as follows. Section II generates a hash code for data records collected from mul-
discusses the existing literature done on application related tiple drones. The system reduces the drone’s communication
to Intelligent IoT with blockchain. Section III describes the burden.
background of blockchain technology and proposed architec- A. Outchakoucht et. al. [19] proposed a blockchain frame-
ture. Section IV discusses the results and comparative analysis. work to provide an effective, scalable, and self-adjusted se-
Section V highlights the conclusion and future work. curity policy with the distributed aspect which is strongly
recommended in the IoT and machine learning algorithms.
II. LITERATURE REVIEW
It ends with an explicit inference system designed for better
K. Kostal and K. Salah [11] used a private blockchain understanding. E. Karafiloski, and A. Mishev [20] emphasizes
for IoT device configuration management system in CISCO the importance of Blockchain technology to some of the Big
networks platforms. It is tested in two ways: firstly, the utility Data areas in providing solutions. The key aspects proposed
of the proposed system design is verified by a functional by the authors are related to user authentication, user-specific
scenario. In the second phase, it is used to estimate download restriction access, data protection and user data record keeping.
time, accept and install time of the updated configuration file. J. Ali et. al. [21] developed a custom Behaviour Monitor
This system was run on CISCO TCL scripts. J. Lee [12] used a which extracts the activity of each device for analyzing their
blockchain patch-transporter, an approach to update software behaviour with the use of machine learning algorithms. The
in IoT Platforms in a peer-peer manner. In this method, a authors also proposed Trusted Execution Technology (TEE)
blockchain based smart contract is used for validating trans- to formulate a secure execution environment for sensitive
portation patches and enticement. P. Helebrandt et. al. [13] application code and data on the blockchain.
introduced private blockchain-based architecture to provide The papers studied in this section along with the key
security in enterprise-level in the creation of the trust system findings and their limitations is summarized in Table I.
among distributed communication parties to resolve security
issues. This private blockchain system is developed to defend III. P ROPOSED B LOCKCHAIN BASED P REDICTION M ODEL
unauthorized data creation and improvise data integrity. The IoT network which is deployed in any real time
C. Qu et. al. [14] proposed an IoT credibility verification environment will continuously sense some required data via
method based on self-organized blockchain structures of layers the sensors. These sensor values are transmitted to its network
and communication parties in order to provide security for var- storage for further decision making. The proposed architecture
ious data-intensive applications running on different machines. is as depicted in the Figure 1 and the steps of the proposed
This method has the advantages of storage space and response model is shown in Algorithm 1.
time. However, the challenge remains for the scenario of large Figure 2 shows the private blockchain framework where
scale IoT platforms to determine how to choose and control the the coastal protection committee will have the authority for
number of blockchain structures (BCS) and the height of the the data block creation and other stakeholders are brought
tree, respectively. U. Guin et. al. [15] developed a blockchain into the blockchain through invites to view the data. Each
architecture based on physically unclonable functions (PUFs) data block contains three values namely: Network id – a
to counter the counterfeit, addressing the clone problems unique number assigned to the set of sensors for identification,
and reliability of the supply chain. This would also provide Battery life prediction – corresponds to the predicted battery
edge device authentication. The PUFs are used to generate life obtained by processing the sensor data through machine
a unique identification number for edge devices registered at learning algorithms [22], Date of predicted value – signifies
manufacturers through hash functions. Registered edge devices the date on which the prediction has occurred. This value can
data will be stored and will be globally available in blockchain be used to determine the time frame for battery exhaustion
 TABLE I
S UMMARY OF SIGNIFICANT WORKS

Ref. Methods Observations Research Gaps

Chain code-CRUD: Create, Read, Improved management Monitoring of Low Accuracy, Moderately secure,
[11] Update, Delete. IoT devices using a private BC
Patch Transporter: Faithful Deliv- For delivering patches through an incen- No specification for architecture, Ac-
[12] ery, Authenticated Origin. tive system curacy.
Introduced architecture for Distributed BC removes single point of Accuracy, Tampering,
[13] monitoring and managing failure
enterprise networks through
private blockchain
SRAM-based physically, Counterfeit, clone problems, reliability Can be used as local block chain only
[15] unclonable functions (PUFs), of supply chain, provides edge device Cannot be scaled
Light Weight, Mining Algorithm authentication.
Smart contact set operations: A proof-of-concept architecture for IoT Accuracy, Tampering.
[16] Revoke permission Query man-
ager Query Permission Remove
manager Add manager
Blockchains, smart contracts Distributed nature build an event-based Less Accuracy, Efficiency. Tampering.
[17] system for managing IoT devices con-
nected to Web of Things gateways

algorithm to get the battery life prediction. This predicted

value will then be uploaded to the blockchain. Any data
uploaded to the blockchain cannot be tampered due to its
immutability.
Hence, the architecture as proposed in Figure 2 would be
very feasible in securing the attained results.
1) Distributed peer-peer network: Blockchain is managed
by a group of people in a distributed peer-peer network. When
a new block is inserted into the chain, a copy of the block is
echoed to all the nodes present in the network. Each node
verifies the correctness of the block and adds it to its chain.
The validity of the block is checked based on the consensus
Fig. 1. Proposed system architecture.
created by the nodes present in the network which tells about
the guidelines, contracts and rules about classifying the block
Algorithm 1: Steps of the proposed model.
as a valid or invalid block. All these mentioned features make
1 Assign network id Initialize sensors network id blockchain more secure than other technologies in the field.
2 for sensor network id do
3 collect(sensor data sd) IV. RESULTS AND DISCUSSIONS
4 if missing sd then The results obtained for the real time beach water based IoT
5 run Attributemean(sd) system in Chicago District are discussed. The methodology
6 end used for predicting the life of the battery in the IoT sensors
7 end and the security provided for storing the predicted battery life
8 Run One-HotEncoding(sd) using blockchain technology is discussed in this section.
9 Convert categorical data to numerical data
10 Run Standardscalar(sd)
11 Normalize sensor data into values between 0 to 1
12 Initialise Predicatedbatterylife bl
13 bl = Run DNN(sd);
14 Store network id, bl, dateofprediction ˜
blockchain;
Fig. 2. Private Blockchain Framework.
15 End;

A. Data Set Description

which in turn can be used to analyze the battery efficiency of The prediction model designed for predicting the battery life
the sensors. of IoT sensors that is deployed in the beach area of Chicago
The set of sensors in the IoT network will generate the District is evaluated using the publicly available data set
data and this data will be processed by the machine learning called ”Chicago Park District Beach Water Quality Automated
 Fig. 5. Predicted Battery Life vs Actual Battery Life.
Fig. 3. Admin side UI for block creation.

to view this data block transacted to the blockchain. The user

side search UI as shown in Figure 4 may be used to obtain
data of sensors in the IoT network by searching the block with
its network ID. A set of all datablock related to that ID will
be shown as a list.

C. Performance Analysis
The proposed prediction model based on DNN is compiled
Fig. 4. Stakeholders side UI. for 7 random instances and the result obtained is plotted in
Figure 5. The battery life of the sensors which is predicted
using the Proposed DNN based Prediction Model for seven
Sensors Data”. There are six sensors that are deployed at the random instances are tabulated in the Table II. When com-
beach for monitoring the impact of waves. Since the sensors pared to the actual battery life, the predicted battery life has
work on batteries, there is a chance that the battery drains an average accuracy of 90%.
out and the data is lost. Hence it is necessary to predict the
life of battery well in advance and report to the Chicago TABLE II
Government based on various attributes and parameters. The C OMPARISON OF ACTUAL V S P REDICTED BATTERY L IFE USING DNN
data set which is used to support this prediction model has
Actual Battery Life Predicted Battery Life
several features like name of th ebeach, water temperature, 60 57.15
turbidity, transduder depth, period of wave, height of the wave, 53 63.34
battery life (class label). 80 60.95
70 57.98
B. Proposed Prediction Model and Results 61 58.30
74 62.39
The prediction model that is proposed is based on Deep 69 63.84
Neural Networks [23], [24]. Deep Neural Networks is much
better as compared to other Machine Learning models due
to the availability of high performance computational systems D. Insight of the Predicted Value
such as Graphical Processing Units. There is one input layer, The prediction model predicts the life of the battery of
five layers that are hidden and finally one layer which forms sensors deployed in the beaches of the Chicago District in
the output. There are a total of 128 neurons in the first six locations and the predicted value will be stored in the
layer which forms the input layer. Since only one output is Blockchain which is publicly available and the government
expected, that is the battery life, one neuron is used for this accesses the data whenever required for replacing the batteries
layer. The hidden layers vary from 64 to 256 neurons. The at the right time as well as to place orders to the corporate
DNN is activated using the Relu Activation Function and companies who manufacture the batteries.
the parameters required for the execution of the model are The prediction accuracy is increased by 12%. The proposed
optimized using the ADAM Optimizer. prediction model is evaluated using the parameters like Mean
The proposed system secures the predicted battery life Absolute Error(MAE), Mean Squared Error(MSE), Root Mean
through the blockchain in a trust based manner. The admin Squared Error(RMSE) and Test Variance Score(TVS). The
updates the records with the help of UI depicted in Figure obtained results are tabulated in the Table 4. The tabulated
3. The predicted battery life along with the network ID and results clearly depict that the MAE is 5.17, MSE is 35.70,
date are updated into the UI. Any stakeholder may be able RMSE is 6.02 and TVS is 0.19 in case of the proposed
DNN Prediction Model which outperforms the other state of V. C ONCLUSION AND F UTURE W ORK
art techniques namely XGBoost and Linear Regression. Any
The research is carried out and evaluated using the Chicago
prediction model is considered to be having high performance
Park District Beach Water Data Set which is collected from
when the variance is low. From the results, it is evident that
a real-time IoT network deployed at six locations on the
the variance score is lesser in DNN when compared to the
beaches. The implementation focuses on predicting the battery
other techniques.
life of the deployed sensors during various time frames by
using the DNN based Prediction Model. The proposed model
TABLE III outperforms the other available Machine Learning Models by
T HE E VALUATION OF P REDICTION M ODELS USING S TANDARD M ETRICS
an increase in prediction accuracy of 12%. This system uses
Model Used Mean Mean Root Test blockchain technology as its backend storage to store the
Ab- Squared Mean Variance
solute Error Squared Score
predicted battery life values. This creates an immutable data
Error Error record that can be viewed by all the stakeholders. If the data
DNN (Pro- 5.17 35.70 6.02 0.19 is made tamper-proof, the biggest concern regarding the trust
posed)
of the delivered data can be mitigated. This combination of
Linear Re- 988.72 1681700169.89 41008.53 -
gression 25876324.98 IoT coupled with blockchain can lead to newer possibilities
XGBoost 5.38 45.23 6.72 0.30 in numerous fields such as supply chain management where
real-time tracking of various parameters may be updated live to
the stakeholders with a surety of untampered data. By reducing
E. Significance of using blockchain human intervention within the existing methodologies, a potent
future for automation may be realized sooner using this
The major challenge of the existing centralized storage is combination. Though blockchain has several advantages, it
that the data is owned by a single entity/stakeholder. This has some unsolved research challenges such as scalability and
entity will have to provide access permission to view the data. delay in transaction processing which maybe addressed by
Here, the data in question which is the battery life predicted future researchers.
values may be falsified to provide an unfair advantage. This
issue arises due to a lack of trust which can be solved through R EFERENCES
the use of the distributed ledger of the blockchain [25]. Here
every stakeholder will obtain a copy of this data. This means [1] C. Iwendi, P. K. R. Maddikunta, T. R. Gadekallu, K. Lakshmanna, A. K.
Bashir, M. D. Piran, “A metaheuristic optimization approach for energy
falsification of this data by one entity is not possible since efficiency in the IoT networks,” Software: Practice and Experience,
the change to be valid, needs to be updated in every ledger 2020, 10.1002/spe.2797.
owned by other entities. This establishes trust [26] over the [2] K. R. Malik, Y. Sam and M. Hussain, “A methodology for real-time
data sustainability in smart city: Towards inferencing and analytics for
current state of data. The data hash of the current block in the big-data,” Sustainable Cities and Society, vol. 39, pp. 548–556, 2018,
blockchain is referred by its succeeding block, which means 10.1016/j.scs.2017.11.031.
any change to the block data can be identified through the [3] P. P. Ray, “A survey of IoT cloud platforms,” Future Comput-
ing and Informatics Journal, vol. 1, no. 1-2, pp. 35–46, 2016,
change in its hash. This provides the basis for immutability in 10.1016/j.fcij.2017.02.001.
the blockchain which makes the system foolproof. [4] M. A. Khan, and K. Salah, “IoT security: Review, blockchain solutions,”
Future Generation Computer Systems, vol. 82, pp. 395–411, 2018,
10.1016/j.future.2017.11.022.
F. Real time use case for the proposed system [5] A. F. A. Rahman, M. Daud, and M. Z. Mohamad, “Securing sensor to
cloud ecosystem using internet of things (iot) security framework,” in
Let us consider an example where remote IoT sensors are Proc. of the International Conference on Internet of things and Cloud
deployed deep into a forest for detection of forest fires. These Computing, 2016, pp. 1–5.
[6] J. Chen, J. Wu, H. Liang, S. Mumtaz, J. Li, K. Konstantin, A. K. Bashir,
senors are equipped with batteries left as stand alone units or and R. Nawaz, “Collaborative Trust Blockchain Based Unbiased Control
as network groups. Once deployed, they process real time data Transfer Mechanism for Industrial Automation,” IEEE Transactions on
and send back alerts regarding the same. If the battery for this Industry Applications, pp. 1–1, 2018, 10.1109/TIA.2019.2959550.
sensor falls short and requires an unpredicted maintenance, [7] J. Chaudhry, A. K. Bashir, A. Hassan, J. Haass, and G. Zheng, “Enabling
Technologies for Post Market Surveillance of Medical Devices: Is
the time taken to replace them would result in loss of crucial Blockchain, SDN or a Legacy Network Best for Post-Market Surveil-
data. The loss of this data could potentially mean a forest lance of Medical Devices?,” IEEE Internet Policy Newsletter, 2018.
fire of an area goes undetected, the result of which could be [8] N. Tariq, M. Asim, F. Al-Obeidat, M. Z. Farooqi, T. Baker, M.
Hammoudeh, and Ghafir, Ibrahim, “The security of big data in fog-
catastrophic. This is such one such example where the result enabled IoT applications including blockchain: a survey,” Sensors, vol.
of unpredicted battery replacement could be hazardous. By 19, no. 8, pp. 1788, 2019, 10.3390/s19081788.
combining this prediction of battery life with blockchain, this [9] U. Khalid, M. Asim, T. Baker, P. C K. Hung, M. A. Tariq, and L.
Rafferty, “A decentralized lightweight blockchain-based authentication
creates an immutable sharable copy of the battery life data mechanism for IoT systems,” Cluster Computing, 2020, 10.1007/s10586-
of all the sensors to the necessary stakeholders with complete 020-03058-6.
surety of the shared data. The proposed system can be applied [10] S. Hakak, W. Z. Khan, G. A. Gilkar, M. Imran, and N. Guizani,
“Securing Smart Cities through Blockchain Technology: Architecture,
in similar scenarios where lack of such prediction mechanism Requirements, and Challenges,” IEEE Network, vol. 34, no. 1, pp. 8–
may result in severe loss during business process. 14, 2020, 10.1109/MNET.001.1900178.
[11] K. Kostal and K. Salah, “Management and monitoring of IoT de-
vices using blockchain,” Sensors, vol. 19, no. 4, pp. 856, 2019,
10.3390/s19040856.
[12] J. Lee, “Patch transporter: Incentivized, decentralized software patch
system for WSN and IoT environments,” Sensors, vol. 12, no. 2, pp.
574, 2019, 10.3390/s18020574.
[13] P. Helebrandt, M. Bellus, M. Ries, I. Kotuliak and V. Khilenko,
“Blockchain Adoption for Monitoring and Management of Enterprise
Networks,” in Proc. of 2018 IEEE 9th Annual Information Technology,
Electronics and Mobile Communication Conference (IEMCON), 2018,
pp. 1221–1225.
[14] C. Qu, M. Tao, J. Zhang, X. Hong and R. Yuan, “Blockchain based
credibility verification method for IoT entities,” Security and Commu-
nication Networks, vol. 2018, 2018, 10.1155/2018/7817614.
[15] U. Guin, P. Cui, and A. Skjellum, “Ensuring proof-of-authenticity
of iot edge de-vices using blockchain technology,” in Proc. of 2018
IEEE International Conference on Internet of Things (iThings) and
IEEE Green Computing and Commu-nications (GreenCom) and IEEE
Cyber, Physical and Social Computing(CPSCom) and IEEE Smart Data
(SmartData), 2018, pp. 1042–1049.
[16] O. Novo, K. Kostal and K. Salah, “Scalable access management in iot us-
ing blockchain: a performance evaluation,” IEEE Internet of Things Jour-
nal, vol. 6, no. 3, pp. 4694–4701, 2018, 10.1109/JIOT.2018.2879679.
[17] N. Fotiou, I. Pittaras, V. A. Siris, S. Voulgaris, and G. C. Polyzos,
“Secure iot access at scale using blockchains and smart contracts,”
in Proc. of 2019 IEEE 20thInternational Symposium on” A World of
Wireless, Mobile and MultimediaNetworks, 2019, pp. 1–6.
[18] X. Liang, J. Zhao, S. Shetty, and D. Li, “Towards data assurance and
resilience in iot using blockchain,” in Proc. of MILCOM 2017-2017
IEEE Military Communi-cations Conference (MILCOM), 2017, pp. 261–
266.
[19] A. Outchakoucht, E. Hamza, and J. P. Leroy, “Dynamic access control
policy based on blockchain and machine learning for the internet of
things,” Int. J.Adv. Comput. Sci. Appl., vol. 8, no. 7, pp. 417–424, 2017,
10.14569/IJACSA.2017.080757.
[20] E. Karafiloski, and A. Mishev, “Blockchain solutions for big data
challenges: A literature review,” in Proc. of IEEE EUROCON 2017-17th
International Conferenceon Smart Technologies, 2017, pp. 763–768.
[21] J. Ali, A. S. Khalid, E. Yafi, S. Musa, and W. Ahmed, “Towards a
secure behavior modeling for iot networks using blockchain,” arXiv
preprintarXiv:2001.01841, 2020, arXiv:2001.01841v1.
[22] S. Bhattacharya, Siva R. K. S., P. K. R. Maddikunta, R. Kaluri, S.
Singh, T. R. Gadekallu, M. Alazab, and U. Tariq, “A Novel PCA-
Firefly based XGBoost classification model for Intrusion Detection
in Networks using GPU,” Electronics, vol. 9, no. 2, pp. 219, 2019,
10.3390/electronics9020219.
[23] T. R. Gadekallu, N. Khare, S. Bhattacharya, S. Singh, P. K. R. Mad-
dikunta, I-H. Ra, and M. Alazab, “Early Detection of Diabetic Retinopa-
thy Using PCA-Firefly Based Deep Learning Model,” Electronics, vol.
9, no. 2, pp. 274, 2020, 10.3390/electronics9020274.
[24] Reddy, T., RM, S. P., Parimala, M., Chowdhary, C. L., Hakak, S., &
Khan, W. Z. (2020). A deep neural networks based model for uninter-
rupted marine environment monitoring. Computer Communications.
[25] Deepa, N., Pham, Q. V., Nguyen, D. C., Bhattacharya, S., Gadekallu,
T. R., Maddikunta, P. K. R., ... & Pathirana, P. N. (2020). A Survey
on Blockchain for Big Data: Approaches, Opportunities, and Future
Directions. arXiv preprint arXiv:2009.00858.
[26] Alazab, M., Layton, R., Broadhurst, R., & Bouhours, B. (2013, Novem-
ber). Malicious spam emails developments and authorship attribution.
In 2013 Fourth Cybercrime and Trustworthy Computing Workshop (pp.
58-68). IEEE.