DataDough 💸

DataDough is a comprehensive analysis of publicly available data sourced from kaggle, conducted by Shreya Sudan. It encompasses a series of rigorous analytical procedures, including data cleansing, exploratory data analysis, data visualizations, hypothesis testing, and predictive analysis.

---

Introduction 🤓

Welcome to DataDough, a data science project focused on exploring the fascinating realm of data and uncovering insights related to salaries of data scientists. In today's data-driven world, understanding the factors that influence data scientists' salaries is crucial for both professionals and organizations alike. This project aims to shed light on the various factors that contribute to the salaries of data scientists and provide valuable insights into this ever-evolving field.

The dataset salary contains data on not only the salaries of data scientists, but also their experience, salary in USD, residence, company's location, and company's size.

This project not only seeks to answer questions about salary trends but also aims to explore the relationships between different variables and their impact on compensation. By examining the interplay between factors like work experience, employment type, and company size, we aim to uncover valuable insights that can aid professionals in negotiating salaries and assist organizations in formulating competitive compensation packages.

To accomplish these objectives, we employ a range of analytical techniques and tools, including data preprocessing, exploratory data analysis, feature engineering, and machine learning algorithms. The project utilizes popular libraries like pandas, scikit-learn, and plotly to perform data manipulation, modeling, and visualization.

Here are the components of our project:

• Part I : Inferential Analysis

  • Data Cleaning

  • Exploratory Data Analysis (EDA)

  • Hypothesis Testing

• Part II : Predictive Analysis

  • Linear Regression Model

    • Framing the Problem

    • Baseline Model

    • Feature Engineering & Final Model

  • Multiclass Classification

    • Random Forest Classifier

    • Decision Trees

• Conclusion

By the end of DataDough, we aim to provide a comprehensive understanding of the salary landscape for data scientists, empowering both individuals and organizations with actionable insights. Whether you are a data scientist seeking to benchmark your salary or an organization looking to attract and retain top talent, this project strives to offer valuable information and assist in data-driven decision-making.

Here is a preview of the salary dataset:

work_year	experience_level	employment_type	job_title	salary	salary_currency	salary_in_usd	employee_residence	remote_ratio	company_location	company_size
2023	SE	FT	Principal Data Scientist	80000	EUR	85847	ES	100	ES	L
2023	MI	CT	ML Engineer	30000	USD	30000	US	100	US	S
2023	MI	CT	ML Engineer	25500	USD	25500	US	100	US	S
2023	SE	FT	Data Scientist	175000	USD	175000	CA	100	CA	M
2023	SE	FT	Data Scientist	120000	USD	120000	CA	100	CA	M

---

Part I : Inferential Analysis 🔬🔬

Data Cleaning 🧹 🧽

After looking at some preliminary statistics for salary, we decided to undertake the following steps to clean our data and prepare it for use:

1. remote_ratio column was divided by 100 to ensure scalability and correctness if the column were to be used as a numeric feature in a model.
2. work_year was converted to string type as it is a categorical feature here and the interpretation of the 'mean' of the work_year in the descriptive statistics is meaningless.

Here are the premliminary descriptive statistics of salary:

	work_year	salary	salary_in_usd	remote_ratio
count	3755	3755	3755	3755
mean	2022.37	190696	137570	46.2716
std	0.691448	671677	63055.6	48.5891
min	2020	6000	5132	0
25%	2022	100000	95000	0
50%	2022	138000	135000	0
75%	2023	180000	175000	100
max	2023	3.04e+07	450000	100

And the first few rows of the cleaned dataset salary

work_year	experience_level	employment_type	job_title	salary	salary_currency	salary_in_usd	employee_residence	remote_ratio	company_location	company_size
2023	SE	FT	Principal Data Scientist	80000	EUR	85847	ES	1	ES	L
2023	MI	CT	ML Engineer	30000	USD	30000	US	1	US	S
2023	MI	CT	ML Engineer	25500	USD	25500	US	1	US	S
2023	SE	FT	Data Scientist	175000	USD	175000	CA	1	CA	M
2023	SE	FT	Data Scientist	120000	USD	120000	CA	1	CA	M

---

EDA 📊

As part of our EDA we explored the relationship and the distribution of the following variables:

1. The Distribution of salary_in_usd

   • Observations: The distribution of salaries (USD) is fairly normal, and slightly right-skewed. It's a unimodal histogram with its peak at $150,000-$155,000

     <iframe src="assets/edaFirst.html" width=800 height=600 frameBorder=0></iframe>

2. The Distribution of salary_in_usd with respect to company_size

   • Observations: We can derive from the overlaid histogram that the medium companies are generally taller than the smaller and the larger companies. This is because the dataset contains more data-points for medium-sized companies. The boxplots accompanying the histograms confirm that the medium sized companies have more outliers and a higher median than the other two groups.

     <iframe src="assets/edaSecond.html" width=800 height=600 frameBorder=0></iframe>

3. The average salary and salary_in_usd as work_year increases

   • Observations: There is a steady increase in the mean salary_in_usd as the years increase. This could be explained by the usual rise in the general price level or inflation. However, the growth of the mean salary is quite peculiar. The salary column has a very high standard deviation, because of the disparities in the foreign ecxchange values of the different currencies. Thus, salary_in_usd provides a more standardized distribution of salaries.

     <iframe src="assets/edaThird.html" width=800 height=600 frameBorder=0></iframe>

---

Hypothesis Testing 👩🏻‍🔬

Null Hypothesis: The distribution of salaries of different company sizes is drawn from the same population, and any differences in the distribution are purely conincidental.

Alternative Hypothesis: The differences in salaries are not purely coincidental

Test Statistic: Means

p-value: 0.0

Conclusion: It is highly unlikely that the difference in means are purely coincidental. Thus, we reject the null hypothesis.

<iframe src="assets/hypothesis.html" width=800 height=600 frameBorder=0></iframe>

---

Part II : Predictive Analysis 🔮🔮

Linear Regression 📈📈

Framing the Problem 🧮 🤔

Prediction Problem: Predict the salaries of data scientists in USD

Response Variable: salary_in_usd

Regressors: work_year, employee_type

Evaluation Metric: R²

---

Baseline Model 🧩

The Baseline Model uses work_year and employee_type as features. Our model procured a R^2^ score 0.01. This is an extremely poor score and it implies that the baseline model cannot quite capture the nature of the data points.

<iframe src="assets/baseline.html" width=800 height=600 frameBorder=0></iframe>

---

Final Model 🏆

Feature Engineering: For the final model, we conducted further EDA and created more data visualizations. Here are the features we either considered or used for the final model:

1. work_year and employment_type: These two categorical variables were one-hot encoded. These variables seem closely related to the salary variable, as salary_in_usd increases with each work_year and the amount of compensation would depend on whether the employee is working full time, part time, or as freelance.

2. company_size: Deriving from the graph below, the size of the company is likely a factor in the compensation earned by its employees. We one-hot encoded this categorical feature as well.

<iframe src="assets/edaSecond.html" width=800 height=600 frameBorder=0></iframe>

3. experience_level: The experience of the individual would assumably play a huge role in deriving their compensation. However, simply one-hot encoding would not be appropriate, as Executive level is more valued than intermediate level or entry level. Thus, we decided to ordinally encode this feature.

<iframe src="assets/exp_lvl.html" width=800 height=600 frameBorder=0></iframe>

4. remote_ratio: From the overlaid histograms and boxplots below, it is apparent that remote_ratio is also closely related to salary_in_usd. This numeric feature was simply passed through the model and was not engineered.

Our final model procured a R^2^ score of 0.2, which is significant rise from the baseline model's score. However, this model can be further improved or maybe another model may fit the data better than Linear Regression. It is crucial to note that a R^2^ score of 0.2 is also significant because the response variable, salary is an economic variable that is affected by several macroeconomic and idustry factors.

<iframe src="assets/final_mdl.html" width=800 height=600 frameBorder=0></iframe>

---

Multiclass Classification 📋

Framing the Problem 🧮 🤔

Prediction Problem: Predict the experience level of data scientists

Response Variable: experience_level

Features: employee_type, company_size

Evaluation Metric : Accuracy

---

Random Forest Classifier 🌴🌴

<iframe src="assets/randomforest.html" width=800 height=600 frameBorder=0></iframe>

---

Decision Trees 🌳🌳

<iframe src="assets/decisiontreebaseline.html" width=800 height=600 frameBorder=0></iframe>

---

Final Classification Model

<iframe src="assets/final_classification.html" width=800 height=600 frameBorder=0></iframe>

---

Conclusion 🏁🏁