Fraud Detection Insights: Transaction Distribution, Class Imbalance, and Amount-Based Analysis

By abhishek.verma75000 · June 2, 2026

An analysis of 1,000 financial transactions reveals a highly imbalanced fraud landscape, with 98.2% legitimate transactions (982 records) and only 1.8%…

Out of 1,000 total transactions, the vast majority are legitimate. Here's the breakdown: Legitimate (Class 0) transactions account for 982 out of 1,000 (98.2%), while Fraudulent (Class 1) transactions make up just 18 out of 1,000 (1.8%). A pie chart and data table have been generated to visualize this distribution clearly.

Legitimate transactions have a slightly higher average amount ($2,580.63) compared to fraudulent transactions ($2,381.92) — a difference of $198.71, meaning fraudulent transactions average about 7.7% less than legitimate ones. Two visualizations were generated: a bar chart comparing average amounts and a box plot showing the full distribution of transaction amounts for each type.

Yes, fraud frequency does vary across different time ranges in the dataset. The analysis divided the Time column into 10 equal-width ranges and measured both the count and rate of fraudulent transactions in each. A total of 18 fraud cases were identified, and their distribution is uneven across the time ranges. Two charts and summary tables were generated to visualize these patterns clearly.

Among features V1–V5, V4 most strongly differentiates fraudulent from legitimate transactions, followed by V1, V3, V2, and V5. Two bar charts and summary tables were generated showing each feature's separation strength and mean values by transaction class.

The amount range with the highest concentration of fraudulent transactions is $3,500–$4,000, which has a fraud rate of 4.63% (5 fraudulent transactions out of 108 total). This range also has the highest absolute count of fraud cases. Two bar charts were generated: one showing fraud rate (%) by amount range, and another showing the raw count of fraudulent transactions per range.

The analysis produced two visualizations to help answer whether high-amount transactions are more likely to be fraudulent. The first bar chart shows the fraud rate (%) broken down by amount quartile (Q1 Low through Q4 High), and the second chart compares the average transaction amount between fraud and non-fraud transactions. These charts reveal how fraud likelihood varies across different spending levels.

The Pearson correlation matrix for V1, V2, V3, V4, V5, and Amount has been calculated and visualized as a heatmap. The analysis reveals that none of the feature pairs exceed the 0.5 correlation threshold — meaning all six features are largely independent of one another. The strongest relationship found is between V1 and V5, with a very weak correlation of r = 0.089.

Using a z-score threshold of z 3 across V1–V5, the analysis flagged 12 out of 1,000 transactions as anomalies. Surprisingly, none of these flagged transactions were fraudulent — all 12 (100%) were legitimate transactions. A grouped bar chart was generated showing the overlap between z-score anomaly flags and actual fraud labels, making it easy to visualize this mismatch.

Detailed summary statistics for V2, V3, and V4 were computed across 982 non-fraud and 18 fraud transactions. The analysis produced comparison tables and bar charts highlighting statistical differences between the two classes. V4 shows the greatest mean difference between fraud and non-fraud transactions, followed by V3 and V2.

Using the IQR method (1.5x rule), the analysis found that no transactions in the Amount column were flagged as outliers. The IQR bounds were calculated as a lower bound of -2,378.20 and an upper bound of 7,586.35, meaning all transaction amounts fall within this range. As a result, 0% of transactions are outliers. The fraud rate among non-outlier transactions (which is all transactions) is 1.80%. Two visualizations were generated: a box plot showing the Amount distribution split by Fraud vs. Non-Fraud class, and a bar chart comparing fraud rates between outlier and non-outlier groups.