What Predicts Fetal Health: A Data-Driven Breakdown

By shrijeetverma13 · July 15, 2026

This analysis identifies which measurements and indicators most reliably distinguish between different fetal health outcomes. By examining correlations,…

The analysis breaks down how records are distributed across the different Class categories in your dataset. A bar chart shows the raw counts for each class, ranked from most to least common, while a pie chart illustrates the proportional share of each class as a percentage of the total. Together, these visuals make it easy to see at a glance which class categories dominate the dataset and which are underrepresented.

The analysis explored how V10 and V12 (variability measures) relate to the Class outcome variable. Three visualizations were generated: box plots showing the distribution of V10 across Class categories, box plots showing V12's distribution across Class categories, and a grouped bar chart comparing the mean values of V10 and V12 for each Class. These charts reveal that the spread and central tendency of both variability measures shift depending on the Class value, suggesting that V10 and V12 do carry information relevant to distinguishing between classes. Looking at the box plots side-by-side lets you see whether one class tends to have wider or narrower spread, and whether the typical (median) values diverge meaningfully between classes.

The analysis compared average values of V2, V3, and V16 across the different Class categories in your dataset. A grouped bar chart and data tables were generated to visually and numerically show how these three variables differ between classes, making it easy to spot which class has the highest or lowest averages for each variable.

Comparing records split at the median Class value (4.0), several histogram statistics show notable differences. V17, V21, V22, and V23 are all lower in the High Class group, while V18 and V24 are higher. The most striking changes are in V17 (dropping from 101.37 to 85.27) and V24 (rising sharply from 8.81 to 29.46), suggesting these two features are the most sensitive to Class level. Data tables summarizing these group statistics were generated for further review.

Among the binary indicators V26-V35, V35 shows the strongest positive relationship with higher Class values, with a correlation of 0.58. When V35 is active, the average Class value jumps to 10.0, more than double the overall average of 4.51. V32 and V34 also show strong positive links, with mean Class values of 7.0 and 9.0 respectively when active. In contrast, V26 and V27 are associated with lower Class values, showing negative correlations of -0.54 and -0.51.

Looking at the top 1% most extreme values for V7 and V11, 44 records stand out from the rest of the dataset. V7 normally ranges from 0 to 564, with the extreme threshold starting at 172, while V11 ranges from 0 to 91 with extremes starting at 78. The single most extreme record had a V7 value of 564 and V11 of 0, and it was associated with Class 6. A scatter plot shows how these extreme points (marked distinctly) compare to the rest of the data across V7 and V11, colored by Class outcome. A bar chart further breaks down how many extreme records fall into each Class category, making it easy to see which outcomes are most associated with these unusual values.

The analysis compared mean values of V6, V8, V13, V25, V31, and V34 between normal (Class 1) and abnormal (Class 2/3) cases to identify which indicators are most elevated in abnormal cases. Tables and a grouped bar chart comparing these columns were generated to support this comparison, though the ranked numerical breakdown could not be fully displayed this time.

Using the IQR method, V9 showed no outliers at all (0 out of the dataset, 0%), V19 had 19 outliers (0.89%), and V24 had the most with 184 outliers (8.65%). This suggests V24 has much more variability and extreme values compared to the other two columns. A chart was generated showing how outlier prevalence differs across each Class value for these three columns, along with detailed summary tables.

The correlation heatmap for the 8 heart rate-related features shows that V4 and V5 have a perfect positive correlation (r = 1.000), meaning they essentially represent the same information. The strongest negative relationship is between V16 and V17 (r = -0.899), indicating that as one increases, the other tends to decrease sharply. These findings suggest V4 and V5 are redundant and one could likely be dropped without losing information, while V16 and V17 move in strongly opposite directions.

Summary statistics (mean, std, min, max, and quartiles) were calculated for V4, V5, V9, V10, V12, and V24 across all 10 fetal health classes. The results reveal substantial differences in both central tendency and spread among classes, and are presented in a comparison table along with charts showing means and error bars for easy visual comparison.