INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)
ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue VIII August 2025
Page 1343
www.rsisinternational.org
Cybercrime Victim Profiling in Nigeria Using Machine Learning and
Psychological Traits
Adole Olotuche Ann*., Benjamin Okike., Dr. Amina Imam
Department of Computer Science University of Abuja, Nigeria.
DOI: https://doi.org/10.51244/IJRSI.2025.120800117
Received: 06 Aug 2025; Accepted: 14 Aug 2025; Published: 12 September 2025
ABSTRACT
Cybercrime victimization is on the rise, yet most existing studies focus on attackers rather than victims. This
research examines the role of psychological traits in predicting cybercrime victimization in Nigeria using
machine learning techniques. The research is motivated by the need to integrate human behavioral factors into
cybersecurity, the study employs Random Forest, Decision Tree, Naïve Bayes, and Logistic Regression
models to analyze thelinks between the Big Five personality traits and victim susceptibility. Data was collected
through a SurveyMonkey questionnaire administered to residents of Abuja Municipal Area Council (AMAC)
and a secondary dataset from an open-access Big Five personality repository. The models were trained and
evaluated using accuracy, precision, recall, and F1 score metrics after data preprocessing. Random Forest
achieved the highest accuracy at 97.2%. From our findings, individuals with high extraversion and low
agreeableness, conscientiousness, emotional stability, and openness are more vulnerable to cybercrime. These
insights support the development of personality-informed cybersecurity awareness and prevention strategies.
Keywords: Cybercrime Victimization, Machine Learning, Big Five Personality Traits, Random Forest,
Psychological Profiling, Nigeria.
INTRODUCTION
Nigerians are increasinglyfalling victim to cybercrime activities because many people are unaware of the
importance of securing their digital information. When people fail to recognize the need for protecting their
sensitive data, the results are mostly devastating due to the fact that such information becomes vulnerable to
breaches which expose the user to a wide array of threats. Consequently, the outcomes frequently include
financial loss, emotional distress, or even identity theft. According to Rauf (2019), home users are particularly
at risk,this is as a result of their low cybersecurity awareness especially when compared to corporate users or
IT professionals. Additionally, these individuals are more present on the internet through prolonged interaction
with social media, which increases their exposure and potential for attack. Therefore, the rise of digital
connectivity in Nigeria while offering numerous benefits, has simultaneously heightened the risk of
cybercrime victimization.
This vulnerability is not only technological in nature. According to Kaakinen et al. (2017), there are also
psychological consequences that vary depending on the individual. The emotional and behavioral responses to
cybercrime differ widely among victims thereby creating a complex pattern of victimization that is not always
visible through technical indicators. These psychological dimensions make it clear that technical defenses
alone are not sufficient to address the growing cyber threat.
The statistical evidence surrounding cybercrime in Nigeria is so troubling. According to WDI (2016),
cybercrime victimization increased from 3.5 percent in 2005 to 47.4 percent in 2014. Alongside this, internet
usage in the country increased dramatically. Alam (2018) observed that mobile phone subscriptions jumped
from just 13.3 per 100 people in 2005 to 82.1 per 100 by 2015. As more Nigerians joined the digital world,
financial losses caused by cybercrimes escalated rapidly. Ogbonnaya (2020) reported that in 2018, Nigeria lost
₦288 billion which is approximately $800 million to various forms of cybercrime. This figure represented a
537 percent increase over the losses recorded in 2017. In the same year, more than 17,600 bank customers
INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)
ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue VIII August 2025
Page 1344
www.rsisinternational.org
reportedly lost around ₦1.9 billion to cyber-related fraud. Projections suggest that by the year 2030, Nigeria
may face losses reaching as high as $6 trillion due to cybercrime thus emphasizing the urgent need for
innovative countermeasures.
The nature of these crimes can be better understood by exploring the two main categories outlined by Weijer
and Leukfeldt (2017). The first category, known as cyber-dependent crimes, refers to offenses that are entirely
reliant on digital technology which include activities such as hacking into protected systems or the deployment
of malicious software that are made possible only through the use of IT infrastructure. This is further explained
by Levi et al. (2017) and Kranenbarg et al. (2019), who emphasize the role of anonymity and technical
capability in facilitating such crimes. The second category, cyber-enabled crimes, involves traditional criminal
activities that are enhanced or scaled through digital platforms which includes the internet fraud, online
harassment, digital stalking, and unauthorized withdrawals from bank accounts. Payne et al. (2019) noted that
the internet allows these crimes to take place more quickly and across wider networks than their offline
equivalents. Rokven et al. (2018) affirmed that while cyber-dependent crimes target information systems
directly, cyber-enabled crimes take advantage of these systems to perpetrate harm more efficiently.
As research into cybersecurity continues to evolve, it is becoming clear that individual psychological
characteristics play a major role in determining vulnerability to cybercrime. One theoretical framework that
provides a useful perspective on this issue is the Big Five personality model, as identified by Weijer and
Leukfeldt (2017). This model outlines five key dimensions of personality: extraversion, agreeableness,
conscientiousness, neuroticism, and openness to experience. Cheng et al. (2020) indicated that people with
high extraversion are sociable and outgoing which makes them more likely to interact with unknown
individuals online. This behavior increases their exposure to potential cyber threats. Similarly, those who score
high on agreeableness are often trusting and cooperative. Sheynovet al. (2023) explained that these individuals
may be more susceptible to phishing attacks or malicious downloads simply because they are more willing to
comply with requests.
Hadlington and Murphy (2018) observed that individuals who demonstrate high conscientiousness are more
structured and cautious, making them less likely to engage in risky online behavior. Equally, people with low
conscientiousness often display forgetfulness and poor decision-making thereby increasing their chances of
being victimized. Albladi et al. (2017) explored neuroticism, which reflects emotional instability. They argued
that highly neurotic individuals tend to be anxious or impulsive making them more prone to falling for scams.
Lastly, openness to experience describes a person’s intellectual curiosity and desire for novelty. According to
Albladi et al. (2017), individuals with low openness are often less engaged in exploratory online activities,
which can reduce their risk of encountering cyber threats.
Abuja, Nigeria's Federal Capital Territory, offers a compelling setting for exploring these issues in greater
detail. Wikipedia.com (2024) reported that Abuja officially became the capital in 1991 and now has a
population of over 1.6 million. The Abuja Municipal Area Council (AMAC), one of six local councils in the
FCT, serves as the focus of this study. With more than 770,000 residents and 12 administrative wards, AMAC
is a rapidly urbanizing region. According to the National Population Commission (2010), and as noted by
Omaojor (2020), crime in Abuja has escalated due to increased internal migration and urban pressures,
including cybercrime incidents.
This study therefore investigates the relationship between the Big Five personality traits and the likelihood of
cybercrime victimization within AMAC. Unlike previous studies that used traditional statistical methods such
as SPSS, this research introduces machine learning techniques including Random Forest, Decision Tree, Naïve
Bayes, and Logistic Regression. The goal is to develop predictive models that can accurately identify which
personality traits correlate most strongly with victimization risk.
This study contributes to both cybersecurity research and practice by bridging the gap between behavioral
psychology and machine learning applications. Its findings offer practical recommendations for the creation of
personality-sensitive awareness programs. These can help educational institutions, law enforcement agencies,
and policymakers design interventions that are not only reactive but also preventive.
INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)
ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue VIII August 2025
Page 1345
www.rsisinternational.org
The remaining part of this paper is organized as follows. Section II presents the background. Section III details
the machine learning methods used in this study. Section IV discusses the results. Finally, Section V provides
the conclusion and outlines possible future research directions.
BACKGROUND AND RELATED LITERATURE REVIEW
Cybercrime has emerged as one of the more unsettling outcomes of widespread digitization. As people spend
more of their lives online (working, shopping, socializing, banking) with the risk of falling victim to digital
crimes grows quietly in the background. Much of the research has focused on understanding the tools and
techniques used by cybercriminals, and while this is necessary, it leaves a significant gap when it comes to
understanding the victims. In particular, the psychological and behavioral factors that might increase
someone's risk of being targeted are still not well understood.
Types and Evolution of Cybercrime
Researchers commonly divide cybercrime into two major categories. The first, known as cyber-dependent
crime, includes offenses that rely entirely on digital technology. These crimes involve activities such as
hacking, the spread of malware, and attacks on information systems (Kranenbarg et al., 2019). The second
category is cyber-enabled crime, which refers to conventional crimes that the internet helps scale or accelerate.
Examples of this type include online fraud, cyberstalking, and identity theft (Weijer&Leukfeldt, 2017; Payne
et al., 2019). Kaur (2018) also distinguishes crimes according to their targets, classifying them as crimes
against individuals, organizations, or digital property.
Beyond their technical classification, the consequences of these crimes can be financially devastating and
emotionally taxing. Hawdon (2021) projected that cybercrime could cost the global economy over $10.5
trillion by 2025. In Nigeria, where internet and mobile adoption has grown rapidly, the financial toll has been
severe. Ogbonnaya (2020) reported that ₦288 billion was lost to cybercrime in 2018 alone. These figures make
it clear that the issue is no longer theoretical or confined to abstract discussions about cybersecurity
infrastructure. Rather, it is a human problem, one that affects real people in tangible ways.
Theoretical Perspectives on Victimization
Various criminological frameworks have been used to understand why certain individuals are more likely to
fall victim to cybercrime. One of the most widely cited is Routine Activity Theory (RAT), developed by
Cohen and Felson in 1979. This theory suggests that crime occurs when a motivated offender meets a suitable
target in the absence of a capable guardian (Andresen & Ha, 2017; Linares, 2014). Although it was initially
used to explain physical-world crimes, researchers have found it relevant for digital spaces as well. For
example, visibility and accessibility in online environments can make someone a more appealing target, just as
walking alone at night might in an offline setting (Leukfeldt&Yar, 2016).
A more behaviorally focused version of this theory is the Lifestyle-Routine Activity Theory (LRAT), which
connectsvictimization to an individual's everyday behavior patterns. According to Herrero et al. (2021), people
who regularly browse the internet late at night, frequently share personal details on social media, or habitually
connect to unsecured networks are more likely to attract cybercriminals. These patterns are especially risky
when combined with low self-control. Self-control theory, introduced by Gottfredson and Hirschi (1990),
posits that people with impulsive tendencies, thrill-seeking behavior, or poor risk assessment are more prone to
victimization (Ngo & Paternoster, 2011; Kwak & Kim, 2022). Such individuals may ignore security warnings
or fall for scams that more cautious users would avoid (Alam, 2018; Nodeland, 2020).
Although these theories provide useful starting points, they often rely on general behavioral indicators and may
overlook the influence of deeper psychological traits.
Personality and Psychology in Cyber Victimization
In recent years, more researchers have begun to explore how personality might shape a person’s vulnerability
to cybercrime. The Big Five personality model has proven useful in this area. This model includes five key
INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)
ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue VIII August 2025
Page 1346
www.rsisinternational.org
traits: extraversion, agreeableness, conscientiousness, emotional stability, and openness to experience
(Weijer&Leukfeldt, 2017; Smith, 2024). These traits are considered relatively stable over time and can
influence behavior in both online and offline contexts.
Individuals high in extraversion are sociable and active on social media, which may increase their exposure to
threats such as phishing or impersonation (Cheng et al., 2020). People who score high in agreeableness are
often trusting and cooperative. Although these are generally positive traits, they may lead someone to comply
with malicious requests more easily (Sheynovet al., 2023). On the other hand, conscientious individuals tend to
be organized and cautious, which can protect them from risky behavior online. Hadlington and Murphy (2018)
found that such individuals are more likely to use strong passwords and avoid suspicious websites.
Emotional stability, often measured inversely as neuroticism, also plays a role. People who are emotionally
unstable may react impulsively, fall for fear-based scams, or make quick decisions without verifying the
source (Albladi& Weir, 2017). Openness to experience, a trait linked to curiosity and imagination, may
encourage exploration of unfamiliar platforms or digital services. While this trait can foster innovation and
learning, it may also increase risk by prompting interactions with untrusted sources (Albladi et al., 2017).
Still, personality traits do not operate in isolation. The same individual may show high openness and high
conscientiousness, creating a more complex behavioral profile. This complexity is something traditional
statistical methods struggle to model effectively.
Previous Empirical Work and Its Limitations
Empirical studies linking personality to cybercrime victimization exist, but most of them suffer from narrow
scopes or methodological constraints. For example, Weijer and Leukfeldt (2017) showed that low
conscientiousness and emotional instability correlated with increased risk of victimization. However, their
study was conducted on a Dutch sample and covered only a limited set of crimes. Abladi and Weir (2017)
reported that four of the Big Five traits influenced susceptibility to cyber-attacks. Their findings were based on
self-reported survey data, which is useful for perception-based studies but may be prone to bias or inaccuracy.
Other studies have focused on adolescents and social behavior. Peluchette et al. (2015) found that extraversion
and openness predicted risky social media usage among teenagers. Peker (2017) identified a similar pattern in
Turkish youth, linking impulsiveness and poor self-control with increased cybercrime exposure. These
findings are valuable, yet many of these studies focus on single traits or do not apply data-driven tools that
could capture interactions across multiple variables.
Even qualitative studies have added nuance. Jensen and Leukfeldt (2018) conducted interviews with victims of
phishing and found that emotional reactions and coping strategies varied widely. Some respondents
experienced long-term anxiety, while others considered the incident minor. These differences suggest that
personality may influence not only the risk of victimization but also how individuals respond after an attack.
Machine Learning for Predicting Victim Profiles
Given the layered and interconnected nature of personality traits, machine learning appears well suited for
analyzing cybercrime victimization. Unlike traditional regression models, machine learning algorithms such as
Random Forest, Naïve Bayes, Decision Trees, and Logistic Regression can process multiple features at once.
This allows them to uncover patterns that might remain hidden in simpler models (Mikkola et al., 2020).
Although few studies have fully embraced this approach, some recent work points in that direction. Herrero et
al. (2021) suggested combining self-control theory and smartphone usage patterns to better understand digital
risk. Akdemir and Christopher (2020) looked at human factors in cybercrime but stopped short of building
predictive models. So far, machine learning has been underused in this space.
This study contributes to closing that gap. By integrating personality data with supervised machine learning
techniques, it aims to move beyond generalizations. The goal is to identify how combinations of traits—rather
INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)
ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue VIII August 2025
Page 1347
www.rsisinternational.org
than isolated characteristics—contribute to a person’s digital vulnerability. In doing so, the study offers not
just academic insight but practical recommendations for cybersecurity awareness and targeted interventions.
METHODOLOGY AND EXPERIMENTAL SETUP
This section outlines the methodological approach and experimental setup used to examine the relationship
between cybercrime victimization and personality traits using machine learning (Figure 1) presents tha visuals.
The goal was to predict victimization susceptibility by analyzing individuals' Big Five personality traits,
leveraging both survey-based primary data and publicly available secondary data. The study was structured to
ensure transparency, replicability, and data-driven rigor.
Data Sources and Preprocessing
Data for this study were drawn from two primary sources. The first was a structured online questionnaire
administered to residents of Abuja Municipal Area Council (AMAC), Nigeria. Participants were 18 years and
older and represented diverse backgrounds, including employed, unemployed, low-income earners, students,
and retirees. The survey was hosted on SurveyMonkey and remained open for a 30-day period. Respondents
were asked to report their experiences with cybercrime victimization and complete items related to the Big
Five personality dimensions.
The second data source was an open-access dataset retrieved from the Kaggle repository, specifically the Open
Psychometrics Project. This secondary dataset consisted of over 700 days' worth of responses to an interactive
online personality test. It contained anonymized records including personality trait scores aligned with the Big
Five framework, along with limited demographic information.
Both datasets underwent preprocessing to prepare for analysis. This included data cleaning, such as handling
missing values and removing incomplete entries. Categorical variables, such as gender, were numerically
encoded (e.g., male = 0, female = 1) to ensure compatibility with machine learning models. The combined
dataset was then normalized to ensure consistent scale across variables. Finally, the full dataset was randomly
partitioned into training and testing subsets using an 80:20 ratio.
Model Architecture and Algorithm Selection
The experimental architecture involved a supervised learning pipeline where victimization categories served as
labels and personality traits (alongside select demographics) were the input features. Four classification
algorithms were selected for their reliability, interpretability, and prior success in behavioral prediction tasks:
Logistic Regression (LR): Used both in traditional statistical analysis and as a machine learning baseline is
used in this study as presented in Equation (1), LR models the probability that a binary outcome variable
󰇝󰇞occurs, given a set of features 󰇛
󰇜 , it offered a benchmark for comparing model
performances and is defined as:
󰇛 󰇜

󰇛



󰇜
(1)
Where:
,
are the feature coefficients, and is Euler’s number (the base of the
natural logarithm).
Naïve Bayes (NB): This probabilistic classifier was chosen for its efficiency on high-dimensional data and
ease of interpretability. Equation (2) NB applies Bayes’ theorem with the “naïve” assumption that all features
are conditionally independent given the class label. The classification rule is presented as (2):
󰇛 󰇜
󰇛󰇜
󰇛
󰇜

󰇛󰇜
(2)
INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)
ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue VIII August 2025
Page 1348
www.rsisinternational.org
Where: 󰇛 󰇜s the posterior probability of class given features, 󰇛󰇜is the prior probability of class
,󰇛
󰇜is the likelihood of feature
given class and 󰇛󰇜is the evidence (often omitted in practice since
it's constant across classes).
Decision Tree (DT): This non-parametric model enabled visual and rule-based insight into how different traits
segmented the population into victim groups. DT, Equation (3) split data based on features that result in the
greatest information gain (or Gini impurity reduction). One common metric is the Gini index, defined for a
node as:
󰇛
󰇜
󰇟
󰇛
󰇜󰇠

(3)
Where: is the number of classes,
󰇛
󰇜
is the proportion of class instances in node . A node is split to
minimize impurity across child nodes.
Random Forest (RF): RF is an ensemble of decision trees, where each tree
outputs a class prediction. This
ensemble method builds multiple decision trees and averages their predictions to improve accuracy and reduce
overfitting. Equation (4) presents how RF final prediction is based on majority voting:
󰇛
󰇛󰇜
󰇛󰇜
󰇛󰇜󰇜(4)
Alternatively:
󰇛 󰇜
󰇛󰇜

Where:
󰇛󰇜is the prediction of the m-th tree, and
󰇛󰇜is the probability of class from tree .
These models were implemented using Python’s scikit-learn library. Prior to training, hyperparameters such as
maximum depth (for Decision Trees) and the number of estimators (for Random Forest) were tuned using
cross-validation on the training data. Default parameters were retained where tuning did not lead to significant
gains.
Experimental Setup and Data Generation
No synthetic data were generated externally. However, augmentation in the form of stratified sampling and
randomized data splits was used to ensure balanced representation of victim categories—cybercrime victims,
traditional crime victims, and non-victims—during training.
The dataset was divided such that 80% was used for training and 20% for model evaluation. All experiments
were run on standard consumer hardware using Python 3.x and Jupyter Notebook environments. Code
execution relied on widely adopted libraries including Pandas, NumPy, Matplotlib, and Seaborn, alongside
scikit-learn.
Evaluation Metrics
To evaluate the performance of each model, several metrics were computed from the test set:
Accuracy: The proportion of correct predictions (both true positives and true negatives)out of total predictions.
It is expressed in Equation (5) as:



(5)
Where: TP = True Positives, TN = True Negatives, FP = False Positives, and FN = False Negatives.
INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)
ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue VIII August 2025
Page 1349
www.rsisinternational.org
Precision: The proportion of true positives among all predicted positives, useful for understanding model
reliability. It is represented as Equation (6):



(6)
Recall: The proportion of true positives correctly identified out of all actual positives, capturing model
sensitivity. Represented as Equation (7):



(7)
F1 Score: The harmonic mean of precision and recall, useful when the classes are imbalanced. Represented as
Equation (8):
 


(8)
Confusion Matrix: Provided a visual breakdown of classification performance across victim types.
Represented as a summarizes prediction results matrix (Equation 9) in a tabular format:
󰇣
 
 
󰇤
(9)
Each element of the matrix represents the count of observations in one of the four categories:TP (correctly
predicted positives), FP (incorrectly predicted positives), FN (incorrectly predicted negatives) and TN
(correctly predicted negatives).
Figure 1: Stepwise Machine Learning Workflow for Cybercrime Victimization Prediction
Reproducibility and Tools
To ensure the study can be replicated, all model-building steps, hyperparameter settings, and preprocessing
procedures were coded using Python. In parallel, traditional logistic regression and multinomial logistic
regression were also conducted using SPSS version 26 to cross-validate key associations. This dual-platform
approach helped verify the consistency and robustness of the results.
RESULT
Understanding cybercrime victimization takes more than identifying who is vulnerable. It also involves asking
deeper questions about how vulnerability manifests and whether we can actually anticipate it in a practical
sense. This section presents the findings from four machine learning models developed to predict cybercrime
victimization based on psychological personality traits and demographic attributes. The analysis is supported
by both statistical outputs and performance metrics, focusing on model accuracy, precision, recall, and F1
score. The results are compared to existing research, with a special focus on the reliability and applicability of
the models in the Nigerian context.
INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)
ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue VIII August 2025
Page 1350
www.rsisinternational.org
Model Performance Overview
We began by assessing each model’s raw performance. The four algorithms tested—Logistic Regression,
Naïve Bayes, Decision Tree, and Random Forest—were evaluated on their ability to classify individuals into
two main categories: traditional crime victims and non-victims. To measure this, we used four key metrics:
accuracy, precision, recall, and F1 score.The results are summarized in Table 4.1.
Table 4.1: Performance Metrics for Each Model
Model
Accuracy
Precision
Recall
F1 Score
Logistic Regression
96.20%
96.40%
96.00%
96.20%
Naïve Bayes
96.10%
96.50%
95.80%
96.10%
Decision Tree
96.50%
96.70%
96.20%
96.40%
Random Forest
97.20%
97.50%
96.90%
97.20%
As shown in Figure 4.1, Random Forest stood out by leading across all four metrics. The margin may appear
modest at first glance; however, even a one percent gain in accuracy becomes significant when applied to
large-scale risk assessments or security screenings. This improvement can mean fewer false alarms and better
targeting of resources.
Figure 4.1: Performance Metrics for Machine Learning Models
The superior performance of Random Forest may be due to the way it builds multiple decision trees on
randomized data subsets and then aggregates their results into a final prediction. This approach reduces both
variance and bias. Consequently, it provides a model that is not only powerful but also less prone to
overfitting.
Interpretation of Model Outputs
Random Forest emerged as the most accurate and reliable model, achieving the highest score across all
metrics. The ensemble structure of Random Forest allowed it to capture complex, nonlinear relationships
between the Big Five traits and victimization classes while minimizing overfitting.
Decision Tree, though slightly behind Random Forest in terms of raw performance, offered valuable
interpretability. By examining tree splits, we identified that conscientiousness and emotional stability
consistently appeared at the top nodes, confirming their relevance as strong predictors of victimization risk.
Logistic Regression and Naïve Bayes produced comparable results and served as effective baseline classifiers.
Although these models lacked the flexibility of tree-based methods, they provided transparent coefficient-
based explanations and reinforced findings from prior research using statistical tools like SPSS.
INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)
ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue VIII August 2025
Page 1351
www.rsisinternational.org
Confusion Matrix Insights
To further understand how well each model performed in classifying individualsthe confusion matrix provides
a more detailed breakdown of predictions. In Figure 4.2, the matrix for the Random Forest model shows how it
performed across the two categories.
The Random Forest model correctly classified 485 out of 500 instances, with only 15 misclassifications. False
positives (Type I errors) and false negatives (Type II errors) were minimal, demonstrating the model’s
precision and generalization strength. Most notably, the false positive rate was 3.2%, and the false negative
rate was 2.8%.
Figure 4.2: Confusion Matrix – Random Forest Model
In total, only 15 out of 500 cases were misclassified. This means the model was correct 97% of the time, with a
recall of 96.9% for traditional crime victims and near-equal specificity for non-victims. It managed to avoid a
strong bias toward either class.
This type of performance is particularly valuable in a real-world security context. If a system fails to recognize
an actual victim, interventions may arrive too late or not at all. On the other hand, mistakenly flagging a non-
victim could lead to unnecessary scrutiny. A model that balances both concerns well is not just accurate—it’s
responsible.
Predictive Value of Personality Traits
Model outputs also provided insights into which personality traits most significantly influenced victimization
risk. Based on feature importance in Random Forest and Decision Tree models, the following hierarchy was
observed:
Low emotional stability (commonly associated with high neuroticism)
Low conscientiousness (linked to disorganization and impulsivity)
Lower levels of agreeableness and openness to experience
Moderate to high extraversion, although its impact was less than expected
These results echoed earlier psychological literature. However, seeing them validated through algorithmic
modeling adds a different dimension. It suggests that behavioral tendencies not only shape personal
interactions but also influence digital vulnerability.
INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)
ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue VIII August 2025
Page 1352
www.rsisinternational.org
This ranking aligns with earlier studies (Albladi& Weir, 2017; Weijer&Leukfeldt, 2017), where traits
associated with risk-aversion (like conscientiousness) and emotional regulation were consistently linked to
lower victimization likelihood. This research presents a number of distinct advantages. Their study used
logistic regression without reporting predictive accuracy or model generalization strength. In contrast, the
current research achieved a 97.2% accuracy rate, offering clear evidence of improved performance using
machine learning techniques.
Moreover, the Dutch-based study did not include diverse geographic or cultural variables. By focusing on
Abuja Municipal Area Council (AMAC), this research integrates localized behavioral and digital access
patterns, providing more culturally nuanced findings. Table 4.2presents a side-by-side comparison.
Table 4.2: Comparison Between Current and Previous Study
Metric
Weijer&Leukfeldt (2017)
Method
Logistic Regression
Context
Netherlands
Accuracy Reported
Not reported
Personality Focus
Big Five traits
Emotional Stability
0.959 (victim)
Conscientiousness
0.981 (victim)
Top Predictors
Emotional Stability, Conscientiousness
Practical Implication
General profiling
When we contrast these results with those of Weijer and Leukfeldt (2017), some differences become
immediately clear. Their study, conducted in the Netherlands, relied solely on multinomial logistic regression
and did not report prediction accuracy. While both studies recognize emotional stability and conscientiousness
as key predictors, our study goes further by quantifying model performance and grounding it in a specific
cultural and regional context. It brings in evidence from a community that’s often underrepresented in digital
security research, especially in Sub-Saharan Africa.Our approach brings in machine learning and applies it to a
Nigerian dataset, specifically residents of Abuja's AMAC area.
DISCUSSION
Trying to predict who might fall victim to cybercrime isn’t a simple task. It goes beyond technical loopholes
and into the human territory, where emotion, behavior, and judgment all play a role. In this study, we took a
behavioral angle, looking into how personality traits might shape someone’s likelihood of becoming a victim.
We also used machine learning to do the heavy lifting in terms of prediction. The goal here is not just to talk
about what worked but to unpack the why behind the results.
How Useful Was the Dataset?
The dataset collected from 500 individuals in Abuja’s Municipal Area Council (AMAC) proved meaningful
for our purpose. It captured not only basic demographics like age and gender but also covered a wide range of
behavioral cues and responses linked to the Big Five personality traits. This created a fuller picture of each
participant. It is likely that this diversity contributed to the models’ strong performance.
Rather than relying on surface-level indicators like income or education, we focused on deeper traits—things
like emotional resilience, conscientiousness, and openness. Pairing this with behavioral questions on internet
INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)
ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue VIII August 2025
Page 1353
www.rsisinternational.org
use and technology access provided useful context. The data was well-balanced across gender and age groups,
which added credibility to the predictive results. Still, one should consider that responses came from self-
reported surveys, which may carry bias. People sometimes paint a better version of themselves. That said, for
this kind of psychological modeling, self-assessment is still a common and accepted practice.
Responding to the Research Questions
RQ1: How can a diverse dataset of cybercrime victims and non-victims be created?
The answer lies in a two-pronged approach. First, a focused local survey, such as the one conducted in AMAC,
can provide demographic and behavioral information that reflects a specific context. Second, merging this with
publicly available datasets, like the open Big Five personality dataset, enhances the depth of personality
coverage. Together, these sources offered a broad and meaningful foundation for victim profiling.
RQ2: How can machine learning algorithms be used to predict cybercrime victimization?
Each of the four models—Logistic Regression, Naïve Bayes, Decision Tree, and Random Forest—was tested
for its predictive strength. Random Forest delivered the most consistent and accurate results across all
evaluation metrics. With an accuracy of 97.2 percent, it slightly outperformed Decision Tree at 96.5 percent,
Logistic Regression at 96.2 percent, and Naïve Bayes at 96.1 percent. This consistency indicates that the
patterns present in the data were meaningful enough for the models to learn from and apply accurately.
False positives and negatives were low, especially in the Random Forest output, where only 15 out of 500
predictions were incorrect. That balance is not just statistically satisfying; it matters in real scenarios where
flagging the wrong person could mean wasted resources or missed threats.
RQ3: Can personality traits predict victimization effectively?
The answer appears to be yes. The models identified low emotional stability and low conscientiousness as the
most predictive traits. These traits are often linked to impulsivity, anxiety, and disorganization—factors that
could increase online vulnerability. Male participants had slightly higher odds of being identified as victims,
and older individuals were somewhat less likely to be flagged. These trends were consistent across multiple
models, especially Naïve Bayes and Logistic Regression.
Interestingly, extraversion—often thought to increase digital risk—did not play as big a role here. It showed up
in the results but didn’t weigh as heavily as emotional stability or conscientiousness. This might suggest that
internal regulatory traits have more to do with victimization than outward sociability, at least in this context.
What Didn’t Quite Match Expectations?
While the models performed well, some findings added unexpected nuance. Extraversion, which many studies
tie to social risk online, wasn’t a leading predictor. That may be because people’s online habits don’t always
match their offline personalities, or it could reflect specific cultural factors in the Nigerian digital space.
Similarly, the Naïve Bayes model, often used as a baseline, held its own against more complex models. This
may suggest that the dataset was especially clean or well-structured, which helped all models succeed.
Why This Matters for Cybersecurity Strategy
If personality traits can be mapped to victimization risk, this opens the door to more personalized
interventions. Training modules and awareness campaigns could be tailored based on individual risk profiles.
For example, people low in conscientiousness might benefit from habit-based digital hygiene training, while
those with low emotional stability might respond better to confidence-building or awareness campaigns that
emphasize emotional control.
Security organizations and educational institutions could use such insights to better support vulnerable
individuals. While ethical safeguards must guide how personality data is used, the potential for early
INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)
ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue VIII August 2025
Page 1354
www.rsisinternational.org
intervention is significant. It’s worth considering how these models might integrate with authentication
systems or awareness tools to provide adaptive support based on risk.
Final Thoughts
So, what does it all mean? First, the data showed clear links between personality and victimization, and the
models picked up on those patterns effectively. Second, the Random Forest model stood out, but all the models
performed better than chance and validated the idea that personality matters in cybersecurity. Third, factors
like gender and age added nuance, with males slightly more at risk and older participants showing marginally
lower susceptibility.
That said, machine learning models are not fortune tellers. They help us see probability, not certainty.
Personality is complex and fluid, and digital behavior often shifts depending on context. These tools should
support decision-making, not replace it. When used responsibly, they can help bridge the gap between
psychological research and cybersecurity practice.
This study doesn’t just offer a high-performing model. It also argues for a shift in how we think about digital
risk—not just in terms of code or clicks but in terms of the people behind the screen.
CONCLUSION
This study set out to explore whether psychological traits could help predict cybercrime victimization using
machine learning. The results offer strong evidence that such an approach is not only feasible but also
effective. Among the models tested, Random Forest proved the most consistent and accurate, achieving a
97.2% success rate. More importantly, the findings suggest that behavioral patterns—especially low emotional
stability and low conscientiousness—play a meaningful role in shaping online vulnerability.
Gender and age emerged as subtle but relevant predictors. Men showed slightly higher susceptibility, while
older participants were somewhat less likely to be classified as victims. These patterns, while not absolute,
reinforce the idea that personality and demographics matter in cybersecurity profiling.
The synthetic dataset, combining survey responses with established personality metrics, demonstrated strong
usability for real-world modeling. Its potential value extends beyond academic curiosity. In both regulatory
and technical contexts, such data could support early-warning systems, personalized cybersecurity education,
and risk-adjusted access protocols.
Looking ahead, these insights may guide the development of personality-informed interventions and targeted
awareness campaigns. As cyber threats continue to evolve, integrating behavioral science into our defense
strategies appears not only useful but necessary. While no model can predict human behavior perfectly, this
research makes a strong case for why we should keep trying.
REFERENCES
1. Akdemir, N., & Lawless, C. J. (2020). Exploring the human factor in cyber-enabled and cyber-
dependent crime victimisation: A lifestyle routine activities approach. Internet Research, 30(6), 1665–
1687.
2. Alam, M. K. (2018). Situational Victimization Among Adolescents: Exploring the Role of Morality,
Self-Control and Lifestyle Risk. Journal of Computer Science, 5(2), 113-130
3. Albladi, S. M., & Weir, G. R. S. (2017). Personality traits and cyber-attack victimisation: Multiple
mediation analysis. 2017 Internet of Things Business Models, Users, and Networks, 6(3), 1–6.
https://doi.org/10.1109/CTTE.2017.8260932.
4. Cheng, C., Chan, L., & Chau, C. (2020). Individual differences in susceptibility to cybercrime
victimization and its psychological aftermath. Computers in Human Behavior, 108, 106311.
https://doi.org/10.1016/j.chb.2020.106311
INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)
ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue VIII August 2025
Page 1355
www.rsisinternational.org
5. Felson, M., & Cohen, L. E. (1980). Human ecology and crime: A routine activity approach. Human
Ecology, 8, 389–406.
6. Hadlington, L., & Murphy, K. (2018). Is Media Multitasking Good for Cybersecurity? Exploring the
Relationship Between Media Multitasking and Everyday Cognitive Failures on Self-Reported Risky
Cybersecurity Behaviors. Cyberpsychology, Behavior, and Social Networking, 21(3), 168–172.
https://doi.org/10.1089/cyber.2017.0524.
7. Hawdon, J. (2021). Cybercrime: Victimization, perpetration, and techniques. American Journal of
Criminal Justice, 46(6), 837–842.
8. Herrero, J., Torres, A., Vivas, P., Hidalgo, A., Rodríguez, F. J., & Urueña, A. (2021). Smartphone
addiction and cybercrime victimization in the context of lifestyles routine activities and self-control
theories: The user’s dual vulnerability model of cybercrime victimization. International Journal of
Environmental Research and Public Health, 18(7), 3763.
9. Hirschi, T. (2004). Self-control and crime. Handbook of Self-Regulation, 537–552.
10. Jansen, J., &Leukfeldt, R. (2018). Coping with cybercrime victimization: An exploratory study into
impact and change. Journal of Qualitative Criminal Justice and Criminology, 6(2), 205–228.
11. Kaakinen, M., Keipi, T., Räsänen, P., & Oksanen, A. (2018). Cybercrime victimization and subjective
well-being: An examination of the buffering effect hypothesis among adolescents and young adults.
Cyberpsychology, Behavior, and Social Networking, 21(2), 129–137.
12. Kaur, E. N. (2018). Introduction of cybercrime and its type. International Research Journal of
Computer Science, 7(4), 71-89.
13. Kwak, H., & Kim, E.-K. (2022). The role of low self-control and risky lifestyles in criminal
victimization: A study of adolescents in South Korea. International Journal of Environmental Research
and Public Health, 19(18), 11500.
14. Leukfeldt, E. R., & Yar, M. (2016). Applying routine activity theory to cybercrime: A theoretical and
empirical analysis. Deviant Behavior, 37(3), 263–280.
15. Kranenbarg, M., Holt, T. J., & Van Gelder, J.-L. (2019). Offending and victimization in the digital age:
Comparing correlates of cybercrime and traditional offending-only, victimization-only and the
victimization-offending overlap. Deviant Behavior, 40(1), 40–55.
16. Levi, M., Doig, A., Gundur, R., Wall, D., & Williams, M. (2017). Cyberfraud and the implications for
effective risk-based responses: themes from UK research. Crime, Law and Social Change, 67, 77–96.
17. Mikkola, M., Oksanen, A., Kaakinen, M., Miller, B. L., Savolainen, I., Sirola, A., Zych, I., & Paek,
H.J. (2020). Situational and individual risk factors for cybercrime victimization in a cross-national
context. International Journal of Offender Therapy and Comparative Criminology,
0306624X20981041.
18. Ngo, F. T., & Paternoster, R. (2011). Cybercrime victimization: An examination of individual and
situational level factors. International Journal of Cyber Criminology, 5(1), 773.
19. Nodeland, B. (2020). The effects of self-control on the cyber victim-offender overlap. International
Journal of Cybersecurity Intelligence & Cybercrime, 3(2), 4–24.
20. Ogbonnaya, M. (2020). Cybercrime in Nigeria demands public-private action-ISS Africa. 2, 6-16.
21. Omaojor, O. (2020). Population Density and Crime Rate in the Federal Capital Territory: The Role of
the Nigerian Police Force [Thesis Dissertation, Nigerian Defence Academy].
Https://Www.Academia.Edu/64117225/Population_Density_And_Crime_Rate_In_The_Federal_Capit
al_Territory_The_Role_Of_The_Nigerian_Police_Force.
22. Payne, B., May, D. C., &Hadzhidimova, L. (2019). America’s most wanted criminals: Comparing
cybercriminals and traditional criminals. Criminal Justice Studies, 32(1), 1–15.
23. Peker, A. (2017). An examination of the relationship between self-control and cyber victimization in
adolescents. Eurasian Journal of Educational Research, 16(67).
24. Peluchette, J. V., Karl, K., Wood, C., & Williams, J. (2015). Cyberbullying victimization: Do victims’
personality and risky social network behaviors contribute to the problem? Computers in Human
Behavior, 52, 424–435.
25. Rauf, A. (2019). The Importance of Human Factor in Cyber security. Journal of Security, 9, 91-98.
26. Rokven, J. J., Weijters, G., Beerthuizen, M. G., & van der Laan, A. M. (2018). Juvenile Delinquency in
the Virtual World: Similarities and Differences between Cyber-Enabled, Cyber-Dependent and Offline
Delinquents in the Netherlands. International Journal of Cyber Criminology, 4(7), 450-479.
INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)
ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue VIII August 2025
Page 1356
www.rsisinternational.org
27. Sheynov, V., Dyatchik, N. & Yermak, V. (2023). Relationships of College Students’ Smartphone
Dependence with Victimization, Vulnerability to Cyberbullying and Manipulations. Pedagogical
Sciences, 5(1), 80-86. Doi:10.52928/2070-1640-2023-39-1-80-86.
28. Smith, T. (2024). Integrated Model of Cybercrime Dynamics: A Comprehensive Model for
Understanding Offending and Victimization in the Digital Realm. International Journal of
Cybersecurity, Intelligence and Cybercrime, 7(2). DOI: https://doi.org/10.52306/2578-3289.1163.
29. Weijer, S. G., &Leukfeldt, E. R. (2017). Big five personality traits of cybercrime victims.
Cyberpsychology, Behavior, and Social Networking, 20(7), 407–412.