Handwriting Personality Recognition with Machine

Learning: A Comparative Study

2020 IEEE 2nd International Conference on Electronics, Control, Optimization and Computer Science (ICECOCS) | 978-1-7281-6921-7/20/$31.00 ©2020 IEEE | DOI: 10.1109/ICECOCS50124.2020.9314529

1st Ahmed Remaida 2nd Aniss Moumen 3rd Younes El Bouzekri El Idrissi
Engineering Science Laboratory Engineering Science Laboratory Engineering Science Laboratory
National School of Applied Sciences National School of Applied Sciences National School of Applied Sciences
Ibn Tofaïl University Ibn Tofaïl University Ibn Tofaïl University
Kenitra, MOROCCO Kenitra, MOROCCO Kenitra, MOROCCO
[email protected] [email protected] [email protected]

4th Benyoussef Abdellaoui

Engineering Science Laboratory
National School of Applied Sciences
Ibn Tofaïl University
Kenitra, MOROCCO
[email protected]

Abstract— Predicting someone’s personality is certainly background surveys and reviews. Then we give a
becoming more and more important in the modern world, a comprehensive description and comparison of the selected
face, voice, handwriting, and many other aspects could be papers, followed by the architecture of our proposed system
personality traits indicators. Investigating Deep Learning and sum up with a conclusion.
algorithms application under this scope could be very benefit
and would provide a solid background for future researches.
This paper aims to presents a comparative study of different II. BACKGROUND STUDIES
approaches and methods using Deep Learning techniques in Going through the literature we noticed that the number
handwriting personality recognition. of works published in the field of computerized personality
detection is growing large interest in the last couple of years.
Keywords— deep learning, artificial neural networks, We also found several surveys and reviews studies focusing
personality analysis, graphology, handwriting analysis. on different approaches of personality detection. Here we
present the surveys and reviews published during the last
I. INTRODUCTION decade that we considered as our background studies.
Personality is a set of classification dimensions (traits) Mehta et al. in 2019[9] reviewed in their work recent
that could represents humans understanding about semantic trends in deep learning based personality detection, they
relationships between adjectives that describes themselves analyzed significant machine learning models which have
from others [1]. The most known and used methods to been employed for personality detection with an emphasis on
measure and model those traits are personality traits models deep learning based methods. By focusing only on
questionnaires such as the Myers Briggs Type Indicator computational approaches without psychological studies on
(MBTI) [2], the Cattell‘s 16PF model [3], the Big-Three personality detection, the authors presents the most popular
model (PEN and EPQ-R) [4] and the Big-Five model [5]. approaches to automated personality detection, followed by
Interpreting personality on the basis of those questionnaires computational datasets, and a state-of-the-art of machine
can be time and effort consuming and could get less accurate learning models for personality detection especially
results when the tested person already knows the purpose of multimodal approaches. They also underlined the need of
the test, for those reasons more researchers are investigating larger, more accurate and more diverse datasets for
alternative methods and ways to measure personality traits personality detection.
like the handwriting personality analysis (Graphology).
Providing a direction for many studies involving
Many handwriting characteristics could give a person‘s computerized handwriting analysis, Jia et al. in 2019[10]
personality, mainly the connection patterns, pressure conducted a survey on Automatic Personality Prediction
patterns, shape of each letter, size of each letter, zone, …etc based on facial image also called Physiognomy by analyzing
[6]. We could refer to Graphologists in order to examine few the prediction results of the existing works. Authors found
individual‘s handwriting samples for personality traits that the big five model get reliable personality traits, leading
prediction, but to analyze hundreds of samples we need to the foundation for the accuracy of the experiments, and that
computerize the process. With the emerging new techniques the prominent problems in the literature are mainly reflected
and algorithms like Deep Learning Artificial Neural in three aspects: Lack of information on research data (all the
Networks, computerized handwriting personality recognition reviewed studies are based on two-dimensional human
is getting more interests in the last decade [7], and more faces), Sampling of the age structure of the sample
accurate systems are built in the process [8]. Due to the lack population (the experimental implementation must be
of comprehensive reviews and studies on handwriting convenient with the authenticity of the data), and Artificial
personality trait recognition using Deep Learning algorithms, design features are not comprehensive (researchers usually
we choose to perform our study in order to review the latest manually design a set of features on the basis of heuristic
trends on this interesting subject. criteria in order to predict personality traits).
Here in this paper, we present a comparative study of Another interesting survey is the one conducted by Ilmini
deep learning models and techniques used in handwriting and Fernand in 2017[11] in which they surveyed the
personality recognition. First, we describe some related

978-1-7281-6921-7/20/$31.00 ©2020 IEEE

Authorized licensed use limited to: Robert Gordon University. Downloaded on May 28,2021 at 22:14:08 UTC from IEEE Xplore. Restrictions apply.
evolution of computational personality assessment and between word feature with 94% accuracy same as dominance
discussed emerging technologies used to identify personality of vertical zone feature, and base line feature with 78%
traits with psychological concepts. Besides providing a solid accuracy.
literature review, authors described methods used to identify
personality traits based on face reading techniques, followed
by deep learning models used for recognition of those traits
from facial features. The authors found that models dealing
with large scale image classification problems such as
GoogLeNet, AlexNet and Microsoft ResNet gave grater
results, also they highlighted that the use of pre-trained
model in new classification task (transfer learning) could be
a great solution for reducing high processing requirement in
convolution neural networks in order to obtain even better
results.
Besides detecting general personality traits, Varshney
and Puri in 2017[12] focused in their survey on forensic Fig. 1. Proposed system architecture for handwriting personality
techniques for personality detection on the basis of recognition [14]
(4 features using multi structure algorithm and baseline feature using multi-
handwriting analysis (also known as graphology) with the layer perceptron)
use of artificial neural networks. The authors aimed to show
a complete methodology for developing a system and to
arouse the concern of developing more efforts in this area. B. Handwriting Analysis for Detection of Personality Traits
As conclusion authors found that a perfectly modeled feature using Machine Learning Approach (Joshi et al., 2015)
extraction system helps to have a better discrimination and a In this work authors presented a novel methodology to
top rank between classes. predict the accurate personality traits of an individual from
baseline, margin, slant of the words, and height of T-bar
III. COMPARATIVE STUDY features of a person's handwriting. In the pre-processing
phase, authors used tresholding to transform gray scale
In this section we aim to make comparisons across
images into binary images, polygonalization to calculate the
different papers presenting approaches and methods for
slant and the baseline, and template matching to single out all
personality recognition on the basis of handwriting and that
the lowercase ‗t‘ and to compare it with the predefined
use deep learning models. We mainly extracted papers
templates. Authors aimed at obtaining personal characteristic
conducting researches on this subject from Springer Link and
traits of individuals especially in the age group of 20-35
IEEE databases and we limited the year of publication to the
years. For that they constructed dataset with feature vectors
past ten years. Here we present a comparison of the works
matrixes of 100 samples of person‘s handwritings within this
done under this scope:
age interval, where the classes are created based on the
characteristics corresponding to particular traits already
A. Application Image Processing to Predict Personality examined and identified by a professional Graphologist. For
Based on Structure of Handwriting and Signature the training set, a back propagation neural networks model is
(Djamal et al., 2013) used to train the dataset. As for the classification, the feature
Based on graphology standards, authors combined vector matrix of a new handwriting sample is crated in order
signature and handwriting for a graphical approach to predict to calculate the similarity with the trained dataset using k-
the personality with the use of multiple structure algorithms nearest-neighbors algorithm, leading to the identification of
and multiple artificial neural networks. In the following we the most appropriate class for this new handwriting sample.
will describe only the handwriting personality recognition
part in the authors‘ system architecture. In this work the
samples used to construct the dataset were collected from 25
different writers as training data and 100 different writers as
testing data. Each of those writers was asked to write a text
and give signature in the box at the bottom. After scanning
the A4 paper in jpg/bmp format, the handwriting samples
were pre-processed and segmented into two handwriting
areas: Signature and Text. The analysis of text handwriting
was carried out on five features: that page margin, spacing
between words, spacing between lines, dominance of vertical
zones, and baselines. For the classification of personality in
relation with the page margin, space between word or line, Fig. 2. The architecture of the novel methodology proposed by the authors
and the dominance of zone features authors used multi [15]
structures algorithms, as for the baseline feature, it was
extracted with Hill and Valley feature and classified with a
Multi-Layer Perceptron (MLP) based on a feed forward C. Study on determining the Myers-Briggs personality type
neural network with back propagation algorithm architecture. based on individual’s handwriting (Gavrilescu, 2015)
The results of handwriting patterns identification based on Authors of this work presented a successfully and
those five features was as the following: page margins and practical system for determining the Myer-Briggs Type
space between line features with 97% accuracy, space Indicators (MBTI) personality types through handwriting,

Authorized licensed use limited to: Robert Gordon University. Downloaded on May 28,2021 at 22:14:08 UTC from IEEE Xplore. Restrictions apply.
dedicated to be implemented in the context of career personality test, a comprehensive database is then created as
counseling and development. The authors conducted their a result of training those inputs-outputs using a multi-layer
research on 64 subjects that they asked to fill the MBTI perceptron neural network (MLP). In the testing stage, the
questionnaire every two weeks for a period of 2 months and created database is used as the main comparison reference,
to provide a writing sample of a predetermined text called the classification was performed with MLP model to identify
―The London letter‖. And to correlate the handwriting the none related properties to the writer and with a Hidden
samples and the MBTI test results authors linked primitive Markov Model (HMM) to identify the related properties to
traits with baseline Graphological standards. As for their the target writer.
system‘s architecture, it was designed on three levels: The
first one is the base level where the handwriting
characteristics and features are determined. The second one
represents the first neural networks layer that determines
eight personality primitives by combining four feed forward
neural networks, the presence or absence as well as the
intensity of those primitives is represented with a set of
percentages and fetched to the second neural network layer
in the next level. The third and last level is where the final
decision is taken, the actual personality type is determined by
combining the separate results provided by the intermediate
level. The author‘s proposed system offers a good accuracy
of (86.7%) in four controlled scenarios with highest
accuracies. By computing the results in less than 1 minute,
authors considered their system to be practical for Fig. 4. The architecture (block diagram) of the proposed personality
determining the MBTI personality traits and to be more recognition system [17]
efficient than the questionnaire form and to be suitable for
real-life use. Authors‘ proposed system achieved a 76% rate in
training phase and 61% rate in testing phase as results of
handwriting personality recognition in comparison to MMPI
test. Authors found that using GDA algorithm in HLAC
feature extraction improved the system‘s performance,
achieving the highest effect (29.08%), while the ratio of
vertical to horizontal words achieved the lowest effect
(8.56%) on the accuracy of the output. Also, they compared
the results obtained with their algorithm with others
algorithms like K-means and K-nearest-neighbors and found
that their proposed system is more accurate than the others.

E. Personality Prediction based on Handwriting using

Machine Learning (Lemos et al., 2018)
Authors of this work proposed a Convolutional Neural
Network (CNN) based personality prediction system that
Fig. 3. Overview of the system‘s architecture proposed by the authors [16] evaluates the handwriting samples on the basis of styles like
baseline (straight, ascending or descending), word spacing
(narrow or wide) and slant (right or left) of the sentence and
D. Identify human personality parameters based on relates those styles to seven personalities characteristics.
handwriting using neural network (Fallah and Their system is divided into four modules: For the first one
Khotanlou, 2016) authors used a website as an input where the image of
Authors in this work presented a personality recognition handwritten text is uploaded by the user. The second module
system from Persian handwritten texts with an innovative in the system is the image pre-processing module, where
method. Their system‘s architecture was based on comparing noise is removed and the image is converted into gray-scale,
personality traits related to extracted features from then the contrast/sharpness of the image is improved to
handwritten text with results of a psychometric test of adult brighten the image, and tresholding is performed to convert
personality and psychopathology used for forensic the image into an array of pixels. In the third module, the
psychology constructed with the Minnesota Multiphasic dataset is created on the basis of the preprocessed images and
Personality Inventory (MMPI). To describe the authors‘ training is performed on this dataset to create a
experiment, 70 individuals selected form ordinary people Convolutional Neutral Network model in order to label the
with different characteristics (age, gender, and education) input images accordingly to five output labels through which
were given the MMPI personality test with 71 fundamental handwritings will be differentiated, each labeled output
questions in first place, after that they were asked to write a represents a specific personality trait found in the
constant text in one paragraph with patience and tranquility. handwriting. In the fourth and last module, the test sample of
Both scores from the test clinical scales and handwriting the handwriting is passed to the CNN model, and as result
graphological scales were used to define the character profile the proposed system gives probabilities in the form of
of the individuals and were stored as results in the dataset. percentages, where the sum of each percentage of personality
For the training stage, the extracted features are considered trait is added up to the total of hundred percent in order to
as inputs while corresponding outputs are created using the predict the most dominant class as mentioned previously.

Authorized licensed use limited to: Robert Gordon University. Downloaded on May 28,2021 at 22:14:08 UTC from IEEE Xplore. Restrictions apply.
 baseline, letter size, ‗t‘ characteristics and the applied pen
pressure. They have extracted features from each of the
above mentioned characteristics from scanned image of
handwriting samples using image processing combined with
existing machine learning approaches in order to determine
the employability of a candidate from a HR perspective. The
dataset consisted of 1890 handwriting samples obtained from
a group of students aged between 20–24 years old; each
sample represents a unique combination of different features
leading to the personality trait that could be desirable or
undesirable from an employer‘s perspective. The
classification process had as inputs the extracted features and
two classes (Yes/No) as outputs where ―Yes‖ means that the
candidate is employable and ―No‖ means the he is ought to
Fig. 5. The architecture (Flow Chart) of the proposed personality be rejected. Authors used different classifiers such as Naïve
prediction system [18] Bayes, Random Forest and Support Vector Machine
and compared them on the basis of their performance. As
results, they found that the SVM classifier had the upper
F. Predicting the Big Five personality traits from hand with a 97% precision accuracy.
handwriting (Gavrilescu and Vizireanu, 2018)
In this work, authors proposed a novel non-invasive
neural network based architecture in order to predict an
individual‘s Big Five personality traits on the basis of his
handwriting. Their study was based on two psychological
measures: The Five Factor Model (FFM) and the
graphological analysis standards. The authors designed their
proposed architecture on three layers: A base layer where the
handwriting sample is normalized and the handwriting
features are acquired, an intermediary layer where a
Handwriting Map is built based on the handwriting features
provided by the base layer, and a top layer where a neural Fig. 7. Process Flow Diagram personality prediction architecture [20]
network is used in order to determine the Big Five
personality type of the writer. As for the construction of the
Dataset, handwritings samples were collected from 128 IV. DISCUSSION
subjects in addition to the FFM test results. Authors tested In the section above we analyzed author‘s works and
their novel architecture following both intra-subject and methodologies and described each of their proposed
inter-subject methodologies; in the intra-subject tests they architecture. In term of accuracy the work of Djamal et al.,
obtained 84.4% accuracy while in inter-subject the prediction 2013 [14] seems to be better than the other works for
accuracy was 80.5%. The highest prediction accuracies were detecting personality traits from page margin, line space,
obtained for Openness to Experience, Neuroticism, and word space and dominance of vertical zones with over 90%
Extraversion, reaching above 84%, while for Agreeableness for similarity classification for those features, though the use
and Conscientiousness they were roughly around 77%. of neural networks was only in detecting personality from
handwriting baseline and got less accuracy (78%). In term of
features used in the process, Fallah and Khotanlou, 2016 [17]
used more features compared to the other works and also
used calculated correlations between those features and
results of the MPPI clinical scales test as inputted data
instead of only using handwriting samples, which explain the
low accuracy of the system but still gives reliable results
when used for clinical purpose. Combining between
reliability and accuracy we could say that both works of
Gavrilescu [16] [19] presents the best architectures for
personality traits detection based on handwriting samples, in
addition to the use of the MBTI and the Five Factor Model
questionnaires to validate the effectiveness, both proposed
Fig. 6. The architecture of the proposed personality prediction system [19] systems computed the results in less than a minute and gives
reliable results for predicting someone‘s personality based on
G. A Machine Learning Approach to Employability his handwriting. Table.1 below presents a detailed
comparison of the seven discussed papers based on features,
Evaluation Using Handwriting Analysis (Joshi et al.,
deep learning algorithms and their accuracies, datasets and
2019) personality traits.
Authors of this work implemented a novel approach for
handwriting analysis for personality traits detection by
focusing on handwriting characteristics such as the margin,

Authorized licensed use limited to: Robert Gordon University. Downloaded on May 28,2021 at 22:14:08 UTC from IEEE Xplore. Restrictions apply.
 TABLE I. COMPARISON OF THE DISCUSSED PAPERS

Deep Learning
Paper title and year Features Accuracy Dataset Personality traits
Algorithms
Page margin
97%
Line space
25 handwriting
Word space samples for training
Dominance of 94% Graphology
Djamal et al., 2013 [14] and
vertical zones Standards
100 samples for
Multi-Layer Perceptron testing
Baseline (MLP) with Back 78%
Propagation Algorithm
Baseline
Page margin Back Propagation
100 handwriting Graphology
Joshi et al., 2015 [15] Neural Network Undeclared
Word slant samples Standards
(BPNN)
T-bar height
Baseline
Pressure Feed Forward Neural Personality traits from
48 handwriting
Strokes Network MBTI test
samples for training
Gavrilescu, 2015 [16] (FFNN) 86.7% +
Lower case ―f‖ and 16 samples for
with Back Propagation Graphology
Lower case ―t‖ testing
Algorithm Standards
Word Slant
V to H ratio
Character size
Page margin
Clinical Scales
Word tilt 70 MMPI test results
From MMPI test
Fallah and Khotanlou, Multi-Layer Perceptron +
Line space 61% +
2016 [17] (MLP) 70 handwriting
Word space Graphology
samples
Standards
Word extension
Line tilt
HLAC
Baseline Convolutional Neural
Graphology
Lemos et al., 2018 [18] Word space Network Undeclared Undeclared
Standards
Slant (CNN)
Baseline
Word Slant
84.4%
Writing 128 * 6
intra- Five Factor Model
Pressure handwriting samples
Feed Forward Neural subject Questionnaire
Gavrilescu and Vizireanu, 2 London‘s letters
Connecting Network (Big Five)
2018 [19] 4 letters (300 word)
Strokes (FFNN) 80.5% +
+
Lower case ‗t‘ inter- Graphology Standards
FFM test results
subject
Lower case ‗f‘
Line Spaces
Margin
Baseline 1890
Support Vector Graphology
Joshi et al., 2019 [20] Letter Size 97% handwriting
Machine (SVM) Standards
Lower case ‗t‘ samples
Pen Pressure

To give more insights to our comparative study we give networks (the choice of the algorithms is still undefined)
our own architecture of a personality traits detection system where each one will estimates a traits of the Big five
based on handwriting. Inspired by all of the discussed works, personality traits. Here we present our handwriting based
we will use more features (handwriting characteristics) to personality traits system‘s architecture:
train a single layer neural network composed of five neural

Authorized licensed use limited to: Robert Gordon University. Downloaded on May 28,2021 at 22:14:08 UTC from IEEE Xplore. Restrictions apply.
 Fig. 8. Our proposed architecture of personality traits recognition system based on the Big Five Model

(ICDAR), Kyoto, Nov. 2017, pp. 621–626, doi:

V. CONCLUSION 10.1109/ICDAR.2017.107.
[9] Y. Mehta, N. Majumder, A. Gelbukh, and E. Cambria, ―Recent trends
In this paper we tried to analyze the potential of deep in deep learning based personality detection,‖ Artificial Intelligence
neural networks when applied to personality prediction based Review, 2019, doi: 10.1007/s10462-019-09770-z.
on handwriting features. We came across of some interesting [10] X. Jia, W. Tian, and Y. Fan, ―Physiognomy in New Era: A Survey of
architectures and approaches for computerized personality Automatic Personality Prediction Based on Facial Image,‖ in IoT as a
Service, vol. 271, B. Li, M. Yang, H. Yuan, and Z. Yan, Eds. Cham:
prediction using Neural Networks. We only analyzed and Springer International Publishing, 2019, pp. 12–29.
compared five architectures due to the limitation of choosing [11] W. M. K. S. Ilmini and T. G. I. Fernando, ―Computational
published papers during the last decade and choosing deep personality traits assessment: A review,‖ in 2017 IEEE International
learning neural networks as techniques, still there are more Conference on Industrial and Information Systems (ICIIS), Dec.
on going works and researches under this scope. We 2017, pp. 1–6, doi: 10.1109/ICIINFS.2017.8300416.
proposed our system‘s architecture for detecting the Big Five [12] A. Varshney and S. Puri, ―A survey on human personality
identification on the basis of handwriting using ANN,‖ in 2017
personality traits from handwriting and intend to explore its International Conference on Inventive Systems and Control (ICISC),
performances in a future work. Coimbatore, India, Jan. 2017, pp. 1–6, doi:
10.1109/ICISC.2017.8068634.
[13] A. Vinciarelli and G. Mohammadi, ―A Survey of Personality
REFERENCES Computing,‖ IEEE Transactions on Affective Computing, vol. 5, no.
[1] I. M. Artha Agastya, D. O. Dwi Handayani, and T. Mantoro, ―A 3, pp. 273–291, Jul. 2014, doi: 10.1109/TAFFC.2014.2330816.
Systematic Literature Review of Deep Learning Algorithms for [14] E. C. Djamal, R. Darmawati, and S. N. Ramdlan, ―Application image
Personality Trait Recognition,‖ in 2019 5th International Conference processing to predict personality based on structure of handwriting
on Computing Engineering and Design (ICCED), Singapore, and signature,‖ in 2013 International Conference on Computer,
Singapore, Apr. 2019, pp. 1–6, doi: Control, Informatics and Its Applications (IC3INA), Jakarta,
10.1109/ICCED46541.2019.9161107. Indonesia, Nov. 2013, pp. 163–168, doi:
[2] D. J. Pittenger, ―Measuring the MBTI... And Coming Up Short,‖ 10.1109/IC3INA.2013.6819167.
Journal of Career Planning and Employment, vol. 54, no. 1, pp. 48– [15] P. Joshi, A. Agarwal, A. Dhavale, R. Suryavanshi, and S. Kodolikar,
52, 1993. ―Handwriting Analysis for Detection of Personality Traits using
[3] H. E. P. Cattell and A. D. Mead, ―The Sixteen Personality Factor Machine Learning Approach,‖ International Journal of Computer
Questionnaire (16PF),‖ in The SAGE Handbook of Personality Applications, vol. 130, no. 15, pp. 40–45, Nov. 2015, doi:
Theory and Assessment: Volume 2 — Personality Measurement and 10.5120/ijca2015907189.
Testing, 1 Oliver‘s Yard, 55 City Road, London EC1Y [16] M. Gavrilescu, ―Study on determining the Myers-Briggs personality
1SP United Kingdom: SAGE Publications Ltd, 2008, pp. 135–159. type based on individual‘s handwriting,‖ in 2015 E-Health and
[4] S. G. Draycott and P. Kline, ―The Big Three or the Big Five—the Bioengineering Conference (EHB), Iasi, Romania, Nov. 2015, pp. 1–
EPQ-R vs the NEO-PI: a research note, replication and elaboration,‖ 6, doi: 10.1109/EHB.2015.7391603.
Personality and Individual Differences, vol. 18, no. 6, pp. 801–804, [17] B. Fallah and H. Khotanlou, ―Identify human personality parameters
Jun. 1995, doi: 10.1016/0191-8869(95)00010-4. based on handwriting using neural network,‖ in 2016 Artificial
[5] G. V. Caprara, C. Barbaranelli, L. Borgogni, and M. Perugini, ―The Intelligence and Robotics (IRANOPEN), Apr. 2016, pp. 120–126,
‗big five questionnaire‘: A new questionnaire to assess the five factor doi: 10.1109/RIOS.2016.7529501.
model,‖ Personality and Individual Differences, vol. 15, no. 3, pp. [18] N. Lemos, K. Shah, R. Rade, and D. Shah, ―Personality Prediction
281–288, Sep. 1993, doi: 10.1016/0191-8869(93)90218-R. based on Handwriting using Machine Learning,‖ in 2018
[6] D. J. Antony and O. F. M. Cap, ―Personality Profile Through International Conference on Computational Techniques, Electronics
Handwriting Analysis,‖ p. 118. and Mechanical Systems (CTEMS), Dec. 2018, pp. 110–113, doi:
[7] A. Remaida, B. Abdellaoui, A. Moumen, and Y. E. B. E. Idrissi, 10.1109/CTEMS.2018.8769221.
―Personality traits analysis using Artificial Neural Networks: A [19] M. Gavrilescu and N. Vizireanu, ―Predicting the Big Five personality
Literature Survey,‖ in 2020 1st International Conference on traits from handwriting,‖ EURASIP Journal on Image and Video
Innovative Research in Applied Science, Engineering and Technology Processing, vol. 2018, no. 1, Dec. 2018, doi: 10.1186/s13640-018-
(IRASET), Apr. 2020, pp. 1–6, doi: 0297-3.
10.1109/IRASET48871.2020.9092076. [20] P. Joshi, P. Ghaskadbi, and S. Tendulkar, ―A Machine Learning
[8] A. H. Garoot, M. Safar, and C. Y. Suen, ―A Comprehensive Survey Approach to Employability Evaluation Using Handwriting Analysis,‖
on Handwriting and Computerized Graphology,‖ in 2017 14th IAPR in Advanced Informatics for Computing Research, vol. 955, A. K.
International Conference on Document Analysis and Recognition Luhach, D. Singh, P.-A. Hsiung, K. B. G. Hawari, P. Lingras, and P.
K. Singh, Eds. Singapore: Springer Singapore, 2019, pp. 253–263.

Authorized licensed use limited to: Robert Gordon University. Downloaded on May 28,2021 at 22:14:08 UTC from IEEE Xplore. Restrictions apply.