Use of text

classification tools
as signatures of
suicidal behavior
Group members:

Abdelrahman Farouk 201700097

Seif Henry 201700262
Rana Ramez 201700548
Yasmin Mabrouk 201701395
Salah El Negoumy 201700223
9/13/2023 Sample Footer Text 1
 Medical problem
According to the WHO approximately 800,000 people die by suicide every year and one of the
major risk factors is mental illness including depression, anxiety, bipolar disorder, substance
abuse, schizophrenia and previous suicide attempts etc..
This AI could be a possible tool in the detection of suicide likelihood; which benefits patients
by providing early intervention and prevention, which will eventually lead to a decrease in
annual suicide rates.
 Gold standard vs the Future

• Gold Standard: Therapy sessions with a licensed psychologist or psychiatrist

who assess the risk of suicide and takes the proper decision according to
the patient's history. This decision can include medications, attending
therapy sessions or getting admitted into a facility. This method has the
disadvantage of human error from the psychiatrist's side and incomplete or
not fully honest history from the patient.
• Our goal for using this AI program is to eliminate both disadvantages
mentioned above by detecting the hidden clues from a person's text or note
to try to determine a more accurate assessment of risk.
• We believe that both methods should be used together to deliver the best
care to those at risk of such a sensitive topic.
 Naïve-Bayes algorithm
The Naive Bayes algorithm works by calculating the probability of a given instance
belonging to a particular class based on the probabilities of its features. It uses a training
dataset to estimate these probabilities, which are then used to classify new instances.
For instance, it has become the de facto standard in text classification for example it can
be used to sort e-mails in either spam or ham sections.
This algorithm was selected because:-
1) it requires relatively few examples for training, but also works well with very large
numbers of examples;
2) it provides the estimated probability for a prediction
3) it is very effective and performs well with noisy data.
 Data input

• This is a text classification study that compared 46 text entries

from the two months before Virginia Woolf’s suicide with 54
texts randomly selected from Virginia Woolf’s work during
other periods of her life. Letters and diaries were included,
while books, novels, short stories, and article fragments were
excluded.
 Analysis and results
• The model showed an accuracy of 80.00% with a kappa index of 0.6
to detect Virginia Woolf’s suicide from the included writings.
• Sensitivity and specificity were 69.23 and 91.67%, respectively, with a
balanced accuracy of 80.45%, controlling for the accuracy of a
random classifier as measured by the expected accuracy.
• The p-value for the text classification model was 0.003, indicating
statistical significance. This suggests that the model's performance is
not due to random chance.
• The positive predictive value (PPV) was
90.00%, indicating that when the model
predicts suicide, it is correct 90% of the
time.
• The negative predictive value (NPV) was
73.33%, meaning that when the model
predicts non-suicide, it is correct 73.33% of
the time.

• The study demonstrates the potential of

using machine learning to identify
linguistic markers associated with suicidal
tendencies in text data. It suggests the
possibility of applying real-time machine
learning to predict unfavorable outcomes
like suicide in individuals with psychiatric
disorders.
• The word cloud analysis showed that certain
words were exclusive to the suicide-related
writings, such as "blue," "books," "house,"
"miss," "she," "suppose," "yes," "you," "Vita,"
and "war." Some common words were more
frequent in non-suicide writings, including
"book," "write," "letter," "can," "come,"
"Leonard," "like," "shall," "must," and "will."
However, "never" and "nothing" appeared
more frequently in non-suicide writings.
 Similar AI programs

1.Decision trees: Decision trees are used to create a model that predicts the value of a
2.Random trees: combines multiple trees to make more accurate predictions
3.support vector machine: it classifies individuals into two different categories based on various features ( it could
classify suicidal people into two groups high risk / low risk )
4.Neural networks : set of algorithms inspired by the human brain, they can learn complex patterns and
relationships in the data making them effective in identifying suicide rates
 Conclusion

• In conclusion, this study demonstrates the potential of text classification

and machine learning in identifying written patterns associated with suicidal
behavior, using Virginia Woolf's diaries and letters as a case study. The
machine learning algorithm exhibited strong performance with a balanced
accuracy of 80.45% sensitivity of 69% and specificity of 91%.
• The integration of artificial intelligence and text analysis opens new avenues
for suicide prevention and mental health support
Other related articles
• Study 1:
- This research aimed to identify suicide attempt risk factors in the general
population using a data-driven machine learning approach. It utilized data from
the National Epidemiologic Survey on Alcohol and Related Conditions (NESARC),
involving more than 2500 survey questions.
- Method: A balanced random forest model was trained using cross-validation to
develop a suicide attempt risk prediction model. Model performance was
assessed based on the area under the receiver operator curve, sensitivity, and
specificity.
- Findings: The study explored risk factors for suicide attempts beyond clinical
populations, considering the general adult population and employing machine
learning techniques for predictive modeling.
- https://pubmed.ncbi.nlm.nih.gov/30085438/
• Study 2:
• This longitudinal study aimed to identify risk factors for future suicide attempts in
individuals with major depressive disorder, using data from the National
Epidemiologic Survey on Alcohol and Related Conditions (NESARC).
• Method: Logistic regression analysis was employed to investigate the associations
between baseline risk factors (including comorbid mental disorders and specific
depressive symptoms) and suicide attempts at a 3-year follow-up.
• Findings: The study found that comorbid anxiety, personality, and substance use
disorders, particularly borderline personality disorder, were strongly associated
with suicide attempts. This suggests that comorbidity of mental illnesses
increases the risk of future suicide attempts in individuals with major depressive
disorder, even after considering their previous suicidal behavior.

• https://pubmed.ncbi.nlm.nih.gov/20122697/
• Study 3:
• This study examines the potential of machine learning in identifying individuals at
risk of suicidal thoughts, attempts, and death within healthcare systems. While
machine learning holds promise for suicide prevention, the study emphasizes the
practical implementation challenges often overlooked. It reviews electronic
health record-based machine learning in suicide research, focusing on crucial
issues like data quality, transparency, and ethics.
• The study acknowledges the difficulty of preventing rare outcomes and suggests
that machine learning could be used for precision medicine in targeting
interventions to specific individuals based on their level of need.
• However, the effectiveness of machine learning for suicide prevention is
hindered by the lack of high-quality, scalable interventions for at-risk individuals
identified through these methods.

• https://pubmed.ncbi.nlm.nih.gov/28866721/
• Study 4:
• In this study, machine learning was used to identify predictors of suicide attempts
among high-risk suicidal soldiers who received outpatient mental health services.
The study compared a group receiving Brief Cognitive Behavioral Therapy for Suicide
Prevention (BCBT) to a group receiving treatment as usual (TAU). Self-report
measures of clinical and demographic variables were analyzed, and the combination
of variables that best predicted suicide attempts was identified.
• The factors that, when combined, correctly classified 30.8% of patients who
attempted suicide during the two-year follow-up included worst-point suicidal
ideation, history of multiple suicide attempts, treatment group (BCBT or TAU),
suicidogenic cognitions, and male sex.
• This combination showed higher sensitivity than many previous models used to
predict suicidal behavior. Overall, the study provides a set of variables that can be
assessed clinically to help identify individuals at high risk of suicide.

• https://pubmed.ncbi.nlm.nih.gov/33113452/
References
• Arsenault-Lapierre G, Kim C, Turecki G. Psychiatric diagnoses in 3275 suicides: a meta-analysis. BMC Psychiatry. 2004;4: 37
10.1186/1471-244X-4-37 - DOI - PMC - PubMed

• Passos IC, Mwangi B, Cao B, Hamilton JE, Wu MJ, Zhang XY, et al. Identifying a clinical signature of suicidality among
patients with mood disorders: A pilot study using a machine learning approach. J Affect Disord. Elsevier; 2016;193: 109–
116. 10.1016/j.jad.2015.12.066 - DOI - PMC - PubMed

• Nordentoft M, Mortensen PB, Pedersen CB. Absolute risk of suicide after first hospital contact in mental disorder. Arch Gen
Psychiatry. 2011;68: 1058–64. 10.1001/archgenpsychiatry.2011.113 - DOI - PubMed

• Pompili M, Gonda X, Serafini G, Innamorati M, Sher L, Amore M, et al. Epidemiology of suicide in bipolar disorders: a
systematic review of the literature. Bipolar Disord. 2013;15: 457–490. 10.1111/bdi.12087 - DOI - PubMed

• Henriksson S, Boethius G, Isacsson G. Suicides are seldom prescribed antidepressants: findings from a prospective
prescription database in Jamtland county, Sweden, 1985–95. Acta Psychiatr Scand. United States; 2001;103: 301–306. -
PubMed
References
• Rozek DC, Andres WC, Smith NB, Leifker FR, Arne K, Jennings G, Dartnell N, Bryan CJ, Rudd MD. Using Machine Learning to Predi ct Suicide Attempts in Military
Personnel. Psychiatry Res. 2020 Dec;294:113515. doi: 10.1016/j.psychres.2020.113515. Epub 2020 Oct 22. PMID: 33113452; PMCID: PMC7719604.
• Rodziński P, Rutkowski K, Ostachowska A. Progression of suicidal ideation to suicidal behavior from a perspective of selected suicidological models. Psychiatr Pol.
2017 Jun 18;51(3):515-530. English, Polish. doi: 10.12740/PP/63984. Epub 2017 Jun 18. PMID: 28866721.
• Bolton JM, Pagura J, Enns MW, Grant B, Sareen J. A population-based longitudinal study of risk factors for suicide attempts in major depressive disorder. J Psychiatr
Res. 2010 Oct;44(13):817-26. doi: 10.1016/j.jpsychires.2010.01.003. Epub 2010 Feb 1. PMID: 20122697; PMCID: PMC2888712.
• Hoertel N, Blanco C, Olfson M, Oquendo MA, Wall MM, Franco S, Leleu H, Lemogne C, Falissard B, Limosin F. A Comprehensive Model of Predictors of Suicide
Attempt in Depressed Individuals and Effect of Treatment-Seeking Behavior: Results From a National 3-Year Prospective Study. J Clin Psychiatry. 2018 Jul
31;79(5):17m11704. doi: 10.4088/JCP.17m11704. PMID: 30085438.