I can help you to..

---

• Harvest data from different sources like APIs, SQL databases or messy Excel spreadsheets.
• Reformat and reshape data to make further analyses easier.
• Clean datasets removing outliers and abberant data.
• Create reproducible data preparation workflows
• Automatize the creation of data related reports

Example:

---

• I collected tweets containing the #surf for 9 months to study how surfers use social media. I used the twitter API and made the data collection automatic thanks to a personnal webserver and a custom R script.
• I created a dashboard for a company called Terres Inovia. It is directly plugged on their SQL database and allows to access its information in real time.

Contact me

---

Tell me more about your project at yan.holtz.data@gmail.com to see if we can work together.

I can help you for

---

• Exploratory analysis of a dataset.
• Creation of clean and reproducible reports using R Markdown.
• Application of statistical methods on your dataset.
• Creation of custom scripts for analysing your data automatically

Example:

---

• I'm the author of several scientific publications where I was in charge to explore and understand complex dataset in order to answer scientific questions.
• I recovered information about crypto currency prices on different exchanges and explored the data looking for arbitrage opportunities.

Contact me

---

Tell me more about your project at yan.holtz.data@gmail.com to see if we can work together.

I can help you with:

---

• Creation and review of graphs for scientific publication.
• Development of data application using R and Shiny.
• Creation of custom data visualization for the web using d3.js.
• Dashboard creation for real time data analysis.
• Visualization of geospatial data.

Example:

---

• A website describing research results on mental disorders comorbidity.
• Data to viz: a decision tree leading to the most appropriate chart type for your data.
• See many more examples in the portfolio section of this website.

Contact me

---

Tell me more about your project at yan.holtz.data@gmail.com to see if we can work together.

Description

---

In this project, we constructed an ML pipeline that predicted changes in the overall severity of depression symptoms in adolescents with high symptoms over a three-month period. We computed the residuals from a linear regression model, predicting the 3-month depression sum score based on the baseline depression sum score. These residuals, which represent the unexplained change over three months, were utilized to build a model that could capture factors not included in the baseline model.

The project entailed comprehensive model building, time-series feature engineering, data preprocessing, model evaluation, and validation to ensure the high quality of input data for the models and the accuracy and interpretability of the results. We trained and evaluated various regression algorithms, including Elastic Net, Random Forest, and Extreme Gradient Boosting, to assess their prediction performance.

Additionally, I visualized the time-series data to gain insights into the trends of eight depressive symptoms (e.g., sadness, energy, interest, etc.).

Read more

---

We are currently drafting the manuscript. To explore the project in detail, please visit the GitHub repository. This repository contains analysis code for this project on the Open Science Framework (OSF).

Description

---

In this project, I developed a machine learning model to predict transaction prices for real estate properties in New York State. Leveraging regression models and extensive data analysis, the goal was to achieve a model with a mean absolute error (MAE) below the project’s win condition of $70,000.

The project involved comprehensive data cleaning, exploratory data analysis, and feature engineering to ensure high-quality input data for the models. Various regression algorithms, including Lasso, Ridge, Elastic Net, Random Forest, and Gradient Boosting Trees, were trained and evaluated for their prediction performance.

Through careful feature engineering and preprocessing techniques, such as standardization and outlier handling, the models demonstrated impressive performance in predicting real estate prices. The Random Forest model emerged as the top-performing model, achieving the highest R2 score and the lowest MAE among the evaluated models.

The project’s outcomes have significant implications for the real estate industry, enabling more accurate and data-driven pricing strategies. The model can assist real estate investment firms in optimizing pricing decisions, assessing risks, and reducing costs associated with manual appraisals. The next steps involve deploying the winning model, continuous monitoring, incorporating external data sources, evaluating model robustness, and collaborating with stakeholders for further improvements.

To explore the project in detail, please visit the ML-real-estate-prediction GitHub repository.

Read more

---

You can read the full report on Github. The source code of the project are on data cleaning notebook and model building notebook.

Description

---

In young and middle-aged adults, suicidal ideation is an important predictor of prospective suicide attempts, but its predictive power in late life remains unclear. In this study, we used Latent Profile Analysis (LPA) in a cohort of depressed older adults to identify distinct ideation profiles and their clinical correlates and test their association with risk of suicidal behavior longitudinally.

This peer-reviewed article has been published on The Journal of Clinical Psychiatry.

As the lead data analyst and the third author of this project, I performed Latent Profile Analysis, Analysis of Variance (ANOVA), and χ2 tests using R. To provide a more comprehensive view of our results, I depicted the clinical characteristics using radar plots and represented long-term changes in suicidal ideation with time-series data spanning 2500 days.

Read more

---

You can read the abstract of the peer-reviewed paper on the Journal of Clinical Psychiatry. Due to copyright restrictions, I'm unable to share the full paper publicly. However, if you're interested in reading the complete paper, please don't hesitate to email me, and I'd be happy to provide you with the published PDF file.

Description

---

In this project, I embarked on the challenge of quantifying discrimination using real-world data. Structural stigma, which encompasses discrimination embedded in societal-level conditions, cultural norms, and institutional policies, has a significant impact on LGBTQ+ individuals in the U.S. While established methods exist to measure individual and interpersonal stigma, quantifying structural stigma poses a unique challenge due to its complex macro-social nature. To address this, I designed an approach using Confirmatory Factor Analysis, utilizing a variety of variables that represented both explicit and implicit attitudes specific to LGBTQ+ individuals from an open data set, Project Implicit. This allowed us to create an index of county-level structural stigma. The resulting factor, when visualized on a U.S. map, provides an illuminating representation of the structural stigma landscape across the country. It helps us better understand the macro-social context and pervasive challenges that LGBTQ+ populations in the U.S. encounter in their day-to-day lives.

Data for this project was sourced from the open-access repository of Project Implicit. We specifically utilized the most recent data, spanning from January to December 2022, with the aim of capturing current societal attitudes towards LGBTQ+. This was achieved by aggregating responses from two separate data sets — Sexuality IAT and Transgender IAT — where participants expressed their implicit and explicit attitudes towards the LGBTQ+ community. Importantly, the data included zip codes provided by the participants, enabling us to link the model-generated factor scores to the corresponding county-level structural stigma.

Read more

---

More details about this project can be found in my blog post. I will update and share the open-source code once the paper is submitted to a journal and published on a preprint website.

Description

---

Gender diverse individuals (i.e., identifying their gender as different from the sex assigned at birth) demonstrate higher rates of non-suicidal self-injury (NSSI) compared to other sexual and gender minority (SGM) populations. Despite the importance of identifying prospective risk factors for NSSI, few studies have examined the longitudinal predictors of NSSI among gender diverse individuals. To address this gap, the current study investigated which factors would prospectively predict past-year non-suicidal self-injurious behaviors among gender diverse individuals.

A supervised machine model based on 10-fold cross-validation of LASSO regularized logistic regressions was built using 31 baseline characteristics. Model features included self-injurious thoughts and behaviors (SITB) related factors, minority stress-related factors, protective factors, mental health factors, and the factors of gender-affirming treatment. Training goal of the model consisted of discriminatory accuracy of presence/absence of NSSI engagement during the 12-month follow-up period.

The model mean cross-validation estimate of the Area Under the Receiving Operating Characteristics Curve suggested an overall good prediction accuracy (AUC, 0.85). Our model suggested the optimal value of tuning parameter “lambda” as 0.03, and selected 8 variables with the strongest association with NSSI: past-year NSSI, past-year NSSI frequency, past-year suicidal ideation, enacted stigma, sense of safety, BSI-GSI, anxiety, and somatization. Lifetime engagement in NSSI, suicidal ideation and suicidal attempts did not emerge as longitudinal predictors in 12-month NSSI.

Read more

---

I can't provide more information on this project for privacy reasons.

Description

---

Objectives
To examine the relationship between childhood traumatic experiences and early and late-onset suicidal behavior among depressed older adults.

Methods
Our sample included 224 adults aged 50+ (M ± SD = 62.5 ± 7.4) recruited into three depressed groups: (1) 84 suicide attempters, (2) 44 suicide ideators, and (3) 58 non-suicidal comparisons, and a non-psychiatric healthy comparison group (N = 38). The Childhood Trauma Questionnaire measured experiences of childhood trauma such as emotional abuse, physical abuse, emotional neglect, physical neglect, and sexual abuse.

Results
Attempters were separated into early- and late-onset based on age of first attempt using a Gaussian mixture model (GMM) that identified a cutoff age of 30 years old.

Overall, we found group differences in emotional and physical abuse and neglect in both genders and sexual abuse in females, but not in males. Early-onset attempters experienced more childhood emotional abuse and neglect than late-onset attempters and were more likely to have experienced multiple forms of abuse. They also experienced more emotional abuse and neglect than all comparison groups. Consistently, early-onset attempters more often met criteria for current or lifetime PTSD relative to late-onset attempters and most comparison groups. Late-onset attempters had similar levels of childhood trauma as other depressed groups.

Conclusions
Our study reaffirms that there are distinct pathways to suicidal behavior in older adults based on their age of first suicide attempt and that trauma experienced in childhood has long-lasting emotional and behavioral consequences, even into late life.

Read more

---

You can read the full text of this peer-reviewed paper on International Psychogeriatrics.

Description

---

The current study investigated the relationship between childhood gender role nonconformity and identity development using baseline data from a multi-site, longitudinal cohort study of US trans and gender nonbinary (TGNB) individuals (N = 330). Less than half of TGNB participants reported being gender role nonconforming in childhood. Participants who reported gender role nonconformity in childhood were more likely to identify as binary trans women or men in late adolescence or adulthood (p < .001), heterosexual (p < .001), and reported higher levels of transgender congruence (p < .001) compared to participants who reported being gender role conforming in childhood. Nonconforming participants reported an earlier age of feeling different (6.7 vs 11.7) and of acknowledging to themselves (13.3 vs 17.7) and to others (19.6 vs 23.9) that they were TGNB; no significant difference was found in the age of changing gender roles, starting gender-affirming hormones and having surgery. In terms of minority stress, the gender role nonconforming group reported lower levels of felt stigma (p < .01) and higher levels of concealment (p < .001). Nonconforming participants were less likely to report depressive symptoms and more likely to report being HIV-positive. Childhood gender role nonconformity appears an important variable affecting TGNB identity development and health. Our findings indicate that TGNB individuals who were gender role nonconforming in childhood reached identity development milestones of awareness and disclosure at an earlier age, and perhaps because of that, reported lower levels of felt stigma and depression symptoms.

Read more

---

Description

---

Objective
Rural teens are less likely to access care for depression than urban teens. Evidence-based digital single-session interventions (SSIs), offered via social media advertisements, may be well-suited to narrowing this gap in treatment access and increasing rural adolescents’ access to support. We evaluated the viability of using social media-based advertisements to equitably recruit rural adolescents with elevated depression symptoms online SSIs; we sought to characterize and assess whether SSI completion rates and acceptability differed for rural versus more urban adolescents, across three intervention conditions (two active, evidence-based SSIs; one placebo control); and we tested whether digital SSIs differentially reduced depressive symptoms.

Method
We used pre-intervention and three-month follow up data from 13- to 16-year-old adolescents (N = 2,322; 88% female; 55% non-Hispanic White) within a web-based randomized control trial of three free, online SSIs (ClinicalTrials.gov identifier: blinded for review) collected eight months into the COVID-19 pandemic in the United States.

Results
Digital SSIs reached rural and urban adolescents at population-congruent rates; however, social media ads resulted in relative underrepresentation of rural youths who holding minoritized racial/ethnic identities. Rural adolescents also completed digital SSIs at similar rates to their urban peers, found SSIs equivalently as acceptable, and reported comparable depression symptom reductions as urban youth.

Conclusion
Digital SSIs and their dissemination through social media may offer a promising means of narrowing the gap between rural and urban adolescents’ access to evidence-based mental health support; however, targeted efforts are warranted to reach racially minoritized youths in rural U.S. counties.

Read more

---

See the associated github repository for more information and source code of the project.

Description

---

Background
Cognitive impairment has been associated with a higher rate of past suicide attempts and a higher likelihood of future re-attempts in older adults, however, the relationship between structural brain alterations, suicidal behavior, and suicide attempt contemplation is unknown.

Method
Participants were 74 adults aged 50+, including suicide attempters (N=25), suicide ideators (N=32), and non-suicidal depressed controls (N=17). Brain volumes were extracted from T1-weighted structural MRI scans using Freesurfer. Onset and severity of depression, as well as lifetime anxiety and substance use disorders, were assessed with the SCID-IV and the Hamilton Depression Rating Scale. Volumetric differences were examined by linear regression analyses in hippocampus, amygdala, thalamus, and entorhinal cortex by study groups and clinical history.

Results
Suicide attempters had significantly smaller right hippocampus (t(70)= 2.713, p=0.008) and left hippocampus volume (t(70)= 2.252, p=0.027) compared to non-suicidal depressed controls. Additionally, attempters showed an earlier onset of depression, as well as more lifetime anxiety disorder, compared to non-suicidal depressed controls. However, amygdala, entorhinal cortex, and thalamus volumes were not significantly different between the three groups (ps > 0.05)

Conclusion
We found an association between hippocampus volume and late-life suicide attempters, who showed greater disease burden marked by an earlier age of depression onset and higher comorbidity. These findings suggest neurodegenerative processes in late-life suicide attempters may be more severe compared to non-suicidal depressed patients. Smaller hippocampus volume could indicate deficits in model-based planning, leading to a poorer internal sense of the consequences of one’s actions, and, in turn, to suicidal behavior

Read more

---

This poster has been presented at 2021 International Summit on Suicide Research, where I was nominated as having one of the top 10 of 190 submitted posters.

Description

---

I created an R script that recovers all tweets containing the hashtags #surf and run it on a webserver for about 9 months. I recovered more than 400k tweets. Some of them being geolocalized for both home and current location, it is possible to deduce where surfers travel.

Here is an overview of the result:

Read more

---

See the dedicated page in data-to-viz.com that tells more about this project, with explanation on how to build a connection map.