Self-starter. Data nerd. Creativity.

Welcome to my portfolio website!

Virgil

The Aeneid

# Dashboarding Mission at L'Oréal: Enhancing Data Accessibility and Decision-Making

## Introduction:

In my role at L'Oréal, I had the opportunity to undertake a challenging and impactful dashboarding mission. This mission aimed to improve data management, accessibility, and decision-making processes within the organization.

## Motivations:



Undertaking this mission to develop a comprehensive dashboarding solution for L'Oréal was driven by my passion for tackling substantial projects from end to end. I was eager to embark on a journey that involved extracting, transforming, and visualizing data to deliver impactful business dashboards. The opportunity to work on a large-scale project, starting from the initial data extraction, through the intricate process of data transformation, and culminating in the creation of insightful dashboards, excited me. I was motivated by the chance to leverage my skills and expertise to make a significant impact on L'Oréal's data management and decision-making capabilities.

## I. Data Migration for Improved Accessibility

### 
Developing and Executing a Comprehensive Data Migration Plan


During this mission, I developed and executed a comprehensive data migration plan to transfer over 1TB of data from the local SharePoint to the headquarters SharePoint. This involved meticulous planning, seamless coordination with various teams, and ensuring a smooth transition. By connecting the headquarters SharePoint to Microsoft Power BI, we achieved improved accessibility and efficiency for business insights. Stakeholders across the organization now had easy access to real-time, reliable data.

## II. Enhancing Data Reliability through Advanced Data Cleansing

### Implementing Advanced Data Cleansing Techniques


To enhance the accuracy and reliability of the insights derived from the data, I implemented advanced data cleansing techniques within Power BI. This involved thorough analysis, identification, and rectification of data errors and inconsistencies. Through my expertise, we achieved an impressive 80% reduction in data errors, significantly improving the reliability of the business insights. Decision-makers could now confidently rely on the data presented in the dashboards to make informed choices.

## III. Collaboration with Department Heads for Customized Dashboards

### 
Identifying Key Metrics and KPIs


Collaborating closely with department heads at L'Oréal was vital to the success of this mission. By working collaboratively, we identified key metrics and KPIs specific to each department's objectives and requirements. This collaboration ensured that the dashboards catered to the unique needs of each department. As a result, we created over 20 customized dashboards that provided actionable insights for decision-making. These dashboards enabled department heads to drive strategic initiatives, optimize processes, and achieve their respective business goals.

# Conclusion:


Through my involvement in this dashboarding mission at L'Oréal, I demonstrated expertise in data migration, data cleansing, and the effective utilization of Microsoft Power BI. The project's success was attributed to collaboration with department heads and the creation of customized dashboards that provided actionable insights for decision-making.

Overall, this mission showcased my ability to transform complex data into meaningful visualizations and valuable insights. By improving data accessibility, reliability, and collaboration, I empowered L'Oréal to make data-driven decisions, drive strategic initiatives, and achieve business objectives effectively.

Here is few exemple of Dashboards that I created.

# **PS : THIS IS MOCK DATA** 



![](/images/Com%20-%20RSE%20V2_page-0001.jpg)![](/images/Energie%20-%20CO2%20V2_page-0001.jpg)![](/images/Energie%20-%20L4TF%20V2_page-0001.jpg)![](/images/Logistique%20V2_page-0004.jpg)![](/images/Service%20aux%20r%C3%A9sidents%20V2_page-0001.jpg)![](/images/Service%20aux%20r%C3%A9sidents%20V2_page-0004.jpg)





Dashboarding project for L'Oréal

Reducing Flood Risks in Belgrade Area through AI Solutions

# Motivation

My motivation for creating an interactive dashboard about life and death in the world using Python, Dash, and Plotly was fueled by a LinkedIn post by Thu Hien Vu that demonstrated the potential of Dash to create visually appealing interactive dashboards. As a lover of data visualization, I was drawn to the tool's flexibility and set out to explore it further.

# Content

The dashboard features a header with the title 'Life and Death Dashboard', followed by 4 cards that highlight key figures for life expectancy for women, life expectancy for men, suicide rate, and fertility index. These maps reveal that life expectancy for women is generally higher than for men, and that the world's average fertility rate is above 2, indicating a rising population.

To delve deeper into the data, I created an interactive globe that allows users to see color-coded key indicators for each country, including life expectancy for men and women, happiness score, and fertility rate. The globe exposes significant disparities between countries and even within continents, and suggests correlations between the four indicators – countries with lower life expectancy rates tend to have lower happiness scores and higher fertility rates.

![](/images/a.png)

To further understand the trends in the dataset, I set up two dynamic graphs that allow users to select between scatter plots and histograms. The left graph examines the relationship between suicide rate and GDP per capita, but finds no link between the two. The right graph, on the other hand, reveals a strong correlation between happiness score and life expectancy, and also shows that fertility rate is correlated to happiness score – countries with higher fertility rates tend to have lower happiness scores.

![](/images/histo-73dc0a69.png)![](/images/graph-714af67b.png)

Then I wanted to look at the relationship between happiness index and life expectancy. Are the happiest people the ones who live the longest? Actually there appears to be a strong correlation between the happiness index and life expectancy. Moreover, we notice that the fertility rate (depending on the size of the circles) is correlated to the happiness index. The higher the fertility rate, the lower the happiness index and vice versa. ![](/images/ferti.png)

I also created a map using a 'natural earth' projection that displays a wider range of criteria, including suicide rate and GDP per capita. This map allows for a more detailed examination of the relationships between these variables.

![](/images/map.png)

Finally, I attempted to answer two main questions: Is having children a cause of suicide? And are the unhappiest people most likely to commit suicide? The results of my study suggest that there is no link between fertility rate and suicide rate, and that it is actually the happiest people who are more likely to commit suicide.

![](/images/Fi2-5d401d65.png)

# Conclusion

Overall, this study has allowed me to showcase various correlations between different factors such as suicide rate and happiness index. These insights would not have been possible without data analysis and visualization. I thoroughly enjoyed creating the interactive globe and map, which enabled me to effectively represent and explore the data.


Visualizing Life and Death on a Global Scale: An Interactive Python Dashboard for Data Science

# Content

My dashboard includes a sidebar that provides context for the study and allows users to parameterize the entire dashboard using a time slider. The slider adjusts the dashboard according to the selected year.

![](/images/S3-cf69205c.png)

One of the standout features of my dashboard is a graph that illustrates CO2 emissions and CO2 emissions per inhabitant across different continents and years (modifiable with the slider). The graph shows that CO2 emissions have only grown exponentially since 1750.

![](/images/1885-2d49e927.png)![](/images/2021.png)

I also created a more detailed graph that shows the sum of CO2 emissions by continent, broken down by source (oil, gas, cement, flaring, coal), and by year (changeable with the slider). This graph reveals that emissions have increased by at least a factor of 10 between 1900 and 2021, and highlights the growing role of Asia as the most emitting continent on the planet.

![](/images/histo2.png)

# Conclusion

In conclusion, my data visualization work on CO2 emissions has brought to light the dominance of industrialized and developed countries in terms of emissions, with the United States, China, and the European Union leading the way. It also shows the steady increase in emissions over time, reflecting the constant demand for energy and fossil fuels. Through this project, I have been able to enhance my data visualization skills and gain a deeper understanding of libraries like Panel and hvplot, allowing me to choose the most suitable options for presenting data.


Exploring the CO2 Emission with an Interactive Python Dashboard using Panel/hvplot

# Motivation

I am passionate about data science and have always been fascinated by the possibilities it offers to understand and analyze data in order to extract useful insights. That's why I decided to undertake a data science project titled 'Analyzing 1000 Data Science Books on Amazon: Insights and Trends'.

By completing this project, I developed new skills in data science, particularly in the areas of text vectorization, text sentiment analysis, and text summary. These skills were particularly useful in analyzing the data to extract the most relevant insights and trends.

This project also allowed me to put my knowledge and skills in data science into practice while learning new techniques and approaches. I was motivated by the opportunity to discover new trends and insights about data science books on Amazon.

I was also motivated by the opportunity to develop my skills in data science and to improve in this discipline. This project offered me the opportunity to work with real data and to put my knowledge and skills acquired through my training and previous experiences into practice.

# Content

In this project, I analyzed 1000 data science books on Amazon in order to uncover the most relevant trends and insights. My primary goal was to answer several key questions:

*   Do more expensive books have better reviews?
*   Do longer books have higher prices?
*   What are the best Python books? What are the best ML books?
*   What is the result of cluster analysis of book names/TF-IDF and K-means?
*   How did I succeed in scraping Amazon reviews and summarizing book reviews?

To answer these questions, I used various data science techniques and approaches, including exploratory data analysis (EDA), text sentiment analysis, cluster analysis, and text vectorization. I also utilized data scraping techniques to gather Amazon reviews and implemented an automated summary of book reviews.
In this article, I will present the results of my project and share my most interesting insights and trends.

## Step 1 - Exploratory Data Analysis

### a- Price vs Review

Do more expensive books have better reviews?![](/images/Capture%20d%E2%80%99%C3%A9cran%202023-01-02%20142510.png)

Based on the scatter plot, there is no clear relationship between the price of a book and its reviews.

### b-Number of page Vs Price

Do longer books have higher prices?

![](/images/Capture%20d%E2%80%99%C3%A9cran%202023-01-02%20143059.png)

In conclusion, it appears that the longest books tend to be the ones that sell for the highest price. This makes sense in terms of content (more content in a larger book) and manufacturing cost. However, it is important to note that this trend may also be influenced by other factors such as the reputation of the author and the demand for the subject matter.

### c-Best Python Books

What is the Top 5 Python books based on number of review and average review score ?![](/images/Capture%20d%E2%80%99%C3%A9cran%202023-01-02%20143742-620a26b3.png)

In conclusion, the following are the top 5 books on Python based on the number of reviews: they have between 2500 and 7000 reviews with an average rating between 4.3 and 4.7. These are really excellent books! I highly recommend them, particularly: *Python Crash Course*

### *d-Cluster Analysis of books*

![](/images/Capture%20d%E2%80%99%C3%A9cran%202023-01-02%20144325.png)

![](/images/Capture%20d%E2%80%99%C3%A9cran%202023-01-02%20144757.png)

For the clustering problem, I used the K-Means Nearest Neighbor method. I first used the elbow method to determine the optimal number of book clusters, and then utilized word clouds to display the most important words within each cluster. This allowed me to visually analyze the distinct characteristics of each cluster and gain a better understanding of the overall trends in the data. Overall, the K-Means Nearest Neighbor method proved to be a useful tool for clustering the data and uncovering insights about the books.

## Step 2 -Scraping Review from Amazon Website

Initially, I noticed a common pattern in the URLs of each book's Amazon link and its review link. Therefore, I created a function to take the URL of the book and convert it into the URL for its review page. This allowed me to easily scrape the reviews for each book and gather the necessary data for my analysis. Overall, this was an important step in my data collection process as it allowed me to efficiently access the review information for all 1000 books.

![](/images/Capture%20d%E2%80%99%C3%A9cran%202023-01-02%20145205.png)

In addition, I utilized Jeff James' code to scrape the reviews for each book and put them into a new data frame. This proved to be a valuable resource as it allowed me to quickly and efficiently gather the review data for all of the books. The use of this code streamlined my data collection process and enabled me to focus on the analysis and interpretation of the data.

![](https://preview--mighty-moon-3465f.stackbit.dev/\_static/app-assets/images/Capture%20d%E2%80%99%C3%A9cran%202023-01-02%20150017.png)

## Step 3 - Sentiment Analysis

### VADER Sentiment Scoring

In addition, I conducted sentiment analysis on each review. I used the VADER model, which is the most basic model for sentiment analysis that has a list of English words with a sentiment score (positive or negative) and calculates the average of all the scores of each word in the text to determine the sentiment of the text. The closer the score is to 1, the more positive the sentiment is, the closer it is to -1, the more negative the sentiment is, and 0 is neutral. This provided valuable insights into the overall sentiment of the reviews for each book and allowed me to analyze trends in the data.![](/images/Capture%20d%E2%80%99%C3%A9cran%202023-01-02%20150452-f426901a.png)

### ROBERTA Sentiment Scoring

urthermore, I used the ROBERTA sentiment scoring model, which takes into account the overall context of the text. This model is more accurate and I compared the results of both models. This allowed me to gain a deeper understanding of the sentiment of the reviews and analyze any discrepancies between the two models. ![](/images/Capture%20d%E2%80%99%C3%A9cran%202023-01-02%20150910.png)

# Conclusion

In summary, I completed a data science project that involved analyzing 1000 data science books on Amazon in order to uncover the most relevant trends and insights. I used various data science techniques and approaches, including exploratory data analysis (EDA), text sentiment analysis, cluster analysis, and text vectorization. I also utilized data scraping techniques to gather Amazon reviews and implemented an automated summary of book reviews.
The results of my project showed that the longest books tend to be the ones that sell for the highest price, although other factors such as the reputation of the author and the demand for the subject matter may also influence this trend. Additionally, I used two different sentiment analysis models, VADER and ROBERTA, to gain a deeper understanding of the sentiment of the reviews and analyze trends in the data. Finally, I used the K-Means Nearest Neighbor method and word clouds to perform cluster analysis and uncover insights about the books. In conclusion, this project allowed me to develop my skills in data science and put my knowledge into practice while learning new techniques and approaches. I am happy to share my results and insights with the data science community in order to contribute to its development.





Analyzing 1000 Data Science Books on Amazon : Insights and Trends

# Motivation

This article analyzes public opinion surrounding a public feud between two YouTubers, KSI and Jake Paul. The two made videos that portrayed each other negatively and their argument became viral. The public had conflicting opinions, with some supporting KSI and others supporting Jake. To understand what people really thought about this feud and whether their opinions changed over time, I scraped data from Twitter and performed a sentiment analysis. In this article, I will guide you through the steps I took to conduct the analysis, including scraping and storing Twitter data, classifying Tweets as positive, negative, or neutral using a simple algorithm, and creating charts with Plotly and Matplotlib to identify trends in sentiment.

# Content

### Step 1 - Data Collection

We will scrape Tweets from the time period when the feud between KSI and Jake occurred, which was in 2020.
I scraped 170K Tweets containing the keywords "Jake Paul" and "KSI" from January 2020.![](/images/Capture%20d%E2%80%99%C3%A9cran%202023-01-02%20135703.png)

Exemple of tweet:![](/images/Capture%20d%E2%80%99%C3%A9cran%202023-01-02%20135752.png)

### Step 2 - Sentiment Analysis

The Tweet mentioned above exhibits negative sentiment. Let's determine if the model can accurately identify this negativity and provide a negative prediction.

The SentimentIntensityAnalyzer (SID) module accepts a string as input and returns scores in four categories: positive, negative, neutral, and compound. The compound score is derived by normalizing the positive, negative, and neutral scores. If the compound score is close to 1, the input string can be classified as having positive sentiment. If it is close to -1, the input string can be classified as having negative sentiment. Let's now use the SID to analyze the sentiment of the above sentence.![](/images/Capture%20d%E2%80%99%C3%A9cran%202023-01-02%20140042.png)

The output of the code above indicates that the sentiment of the input string is negative, with a score of -0.25. This is correct. Now, let's create a function that predicts the sentiment of each Tweet in the dataframe and stores the result as a separate column called "sentiment." Before we do that, we need to run some lines of code to clean the Tweets in the dataframe.
To clean the Tweets in the dataframe, we need to remove the "@" symbol from mentions, eliminate HTTP links, and remove the hashtag symbol "#".

Now that the Tweets have been cleaned, we can perform the sentiment analysis. We need to convert the compound scores into categories — ‘positive’, ‘negative’, and ‘neutral'![](/images/Capture%20d%E2%80%99%C3%A9cran%202023-01-02%20140148-f97c561e.png)

### Step 3 - Visualisation

Now that we have classified the Tweets as positive or negative, we can examine changes in sentiment over time. To do this, we need to group the positive and negative sentiment and count them by date. Then, we can visualize sentiment by date using Plotly.![](/images/Capture%20d%E2%80%99%C3%A9cran%202023-01-02%20140726-140ff2ad.png)

During the period of July-August 2022, there was a noticeable increase in both positive and negative sentiment. This can be seen on the graph where the red line (representing negative sentiment) and the green line (representing positive sentiment) both show a spike. This period coincides with the announcement of Jake Paul proposing to KSI to fight in a boxing match. If we focus our attention on this specific time frame, we can examine the sentiment breakdown of this feud in greater detail.

On August 28, 2022, there was a significant increase in both negative and positive sentiment, as indicated by a spike on the graph. On this particular day, the number of tweets was ten times higher than usual. This can be attributed to the boxing match that took place on this date. The coincidence confirms that our model is accurate. Same for August 8th, which is the fight annoncement.![](/images/Capture%20d%E2%80%99%C3%A9cran%202023-01-02%20140726.png)

Now, let's examine some word clouds to see what people are saying about the event. First, let's look at the most commonly used positive words.

The positive words about Jake and KSI feud and fight are : 'biggest', 'idol', 'interesting', 'hope'. Nice words to describe this fight.![](/images/Capture%20d%E2%80%99%C3%A9cran%202023-01-02%20140931.png)

The negative comments seems to be more aggresive about the result of the fight : 'washed', 'ko', 'beat'![](/images/Capture%20d%E2%80%99%C3%A9cran%202023-01-02%20141044.png)

# Conclusion

In conclusion, the sentiment analysis project on Python involving the scrapping of tweets about the KSI vs Jake Paul boxing match was a success. By analyzing the sentiment of tweets about the event, we were able to gain insight into the public's reactions and opinions on the fight. The results of the analysis showed a spike in both positive and negative sentiment around the time of the match, which is expected given the high level of public interest in the event


Welcome to my portfolio website!

Recent Projects

Dashboarding project for L'Oréal

Reducing Flood Risks in Belgrade Area through AI Solutions

Visualizing Life and Death on a Global Scale: An Interactive Python Dashboard for Data Science

Exploring the CO2 Emission with an Interactive Python Dashboard using Panel/hvplot

Analyzing 1000 Data Science Books on Amazon : Insights and Trends

Python-Powered Twitter Sentiment Analysis of the KSI vs. Jake Paul UFC Fight

A great travel enthusiast.

Home

Welcome to my portfolio website!

Recent Projects

Dashboarding project for L'Oréal

Reducing Flood Risks in Belgrade Area through AI Solutions

Visualizing Life and Death on a Global Scale: An Interactive Python Dashboard for Data Science

Exploring the CO2 Emission with an Interactive Python Dashboard using Panel/hvplot

Analyzing 1000 Data Science Books on Amazon : Insights and Trends

Python-Powered Twitter Sentiment Analysis of the KSI vs. Jake Paul UFC Fight

A great travel enthusiast.