aurele1/diamonds-eda-aurele

README - Diamonds Dataset & EDA Analysis

Dataset Overview + Data Cleaning: The diamonds dataset contains detailed information on 53,940 diamonds, including both numerical and categorical features commonly used in the gemstone and jewelry industry. It is widely used in data analysis and machine learning to study how characteristics such as carat, cut, color, clarity, depth, table, and physical dimensions (x, y, z) influence price. To prepare the dataset for analysis, I first verified that it contains no missing values and no duplicate rows. I then removed an irrelevant column created during CSV export (“Unnamed: 0”), as well as entries with impossible physical dimensions (x, y, or z equal to zero), since a real diamond cannot have a dimension of zero.

Outlier Detection & Decisions: After removing invalid rows, I examined the numerical features to identify potential outliers. Although extreme values were found in variables such as price, carat, and the physical dimensions, these values represent rare but valid diamonds that are larger, heavier, or significantly more expensive than typical ones. Since these observations reflect meaningful real-world variation and contribute important information about diamond pricing, I decided to keep all outliers in the dataset.

Descriptive Statistics

To better understand the distribution of the numerical features in the dataset, I analyzed summary statistics such as the mean, median, standard deviation, and value ranges. The results show large variability in price and carat, with price ranging from $326 to $18,823 and carat ranging from 0.2 to 5.01. These wide ranges indicate strong differences in diamond size and cost across the dataset. In contrast, features like depth and table are more consistent, showing lower variability and suggesting stable manufacturing or cutting proportions. The physical dimensions (x, y, z) also show moderate variation, reflecting differences in diamond shapes and proportions. Overall, these descriptive statistics provide an initial understanding of the dataset and help indicate which features may have a stronger influence on price.

Visualizations

To better understand how the numerical features in the dataset are distributed, I created visualizations such as histograms and a correlation heatmap. These visualizations help reveal patterns such as skewness, variability, and relationships between key variables.

Main Observations (Histograms): -Price is strongly right-skewed, meaning most diamonds are relatively inexpensive while a smaller number reach very high prices. -Carat shows a similar right-skew, indicating that larger diamonds are rare. -Depth and table are more normally distributed, suggesting consistent cutting standards. -x, y, z (dimensions) show moderate spread with a few unusually large values. These insights help highlight which features may require transformation or special handling in later modeling steps.

Correlation Heatmap

The heatmap below shows the correlations between the numerical features in the dataset. The heatmap shows strong positive correlations between carat, price, and the physical dimensions (x, y, z). This means larger diamonds tend to have higher prices and larger measurements. Depth and table show weak or negative correlations with most features, indicating they have less direct influence on price.

Research Questions:

In this section, I explored three research questions to understand how different diamond characteristics influence price. For each question, I used visualizations to compare average prices across categories and identify clear patterns.

Q1: Does cut quality affect price?

This chart shows the average diamond price across different cut categories. Premium diamonds have the highest average price, while Ideal diamonds have the lowest. This indicates that cut quality alone does not determine price. The results suggest that other factors, such as carat size, strongly influence price and may vary across cut categories. Therefore, the relationship between cut and price is not linear and cannot be interpreted independently of other features.

Q2: Does color grade influence price?

This chart shows how the average diamond price varies across color grades. The trend is clear: diamonds with lower color grades (I–J) have the highest average prices, while diamonds with higher color grades (D–F) are less expensive. This is the opposite of what we might expect in theory, where whiter diamonds (D–F) are considered more valuable. This pattern suggests that color alone does not determine price. It is likely influenced by additional factors such as carat size — for example, color grades I and J may include larger diamonds on average, raising their mean price. Therefore, the relationship between color and price in this dataset should be interpreted together with other features, not independently.

Q3: Does clarity impact price?

This chart shows that the relationship between clarity and price is not straightforward. While we might expect higher clarity grades (such as FL, IF, VVS1, VVS2) to be more expensive, the plot reveals that SI2 has the highest average price. This suggests that clarity alone does not determine the price — and that other factors, especially carat size, strongly influence the results. Diamonds with lower clarity can still be more expensive if they are larger.

Summary of Key Insights

The exploratory analysis reveals that diamond price is influenced by multiple factors, but not always in a straightforward or expected way. While cut, color, and clarity all show some relationship with price, the visualizations demonstrate that these attributes do not determine the price on their own. Instead, carat size has a much stronger impact, and it often explains why certain categories with lower theoretical quality (such as SI2 or lower color grades) still show higher average prices. Overall, the findings suggest that diamond size is the main driver of price, while the qualitative attributes (cut, color, clarity) contribute to price variation but only together with carat.

Presentation Video

You can watch the presentation video here: https://huggingface.co/datasets/aurele1/diamonds-eda-aurele/resolve/main/brief%20video.MOV