Fraud Analysis with Random Forest

<b>Title:</b> Fraud Analysis with Random Forest<b>Author:</b> Maria Corina Acosta Araujo<br><b>Context: </b>This study aims to understand which are the most influential factors of fraudulent transactions. The analysis of a critical topic like fraud is important to prevent this crime, to ensure clients’ safety and privacy, and the bank's transparency and profitability. For this purpose, the Random Forests technique is appropriate and useful, as it handles non-linear relationships, it is fitting for classification problems, and it can highlight the most important factors that indicate fraud.<b>Research Question:</b> Is it possible to identify the key features or characteristics of fraudulent transactions using a Random Forest model?<br><b>Hypothesis</b>: <b>Null hypothesis</b>-. The variables aren't statistically significant to prevent fraud.<b>Alternate Hypothesis</b>-. The variables are statistically significant to predict the fraud.<br><b>Data: </b>The dataset used for this project includes a target variable that classifies transactions in fraudulent or not fraudulent, and other predictor variables like amount of the transaction, time, location, etc.<br><b>Data Gathering: </b>The<i> </i>data<b> </b>was collected through Kaggle’s library. No primary source methodology to collect data was implemented.The data set used is the “<b>Credit Card Fraud Detection Dataset 2023</b>” by Elgiriyewithana, (2023).“The dataset includes various anonymized transaction attributes, such as time, location, merchant category, and transaction type. Each transaction is labeled as either fraudulent (1) or non-fraudulent (0), making it an ideal dataset for training and evaluating fraud detection systems” (Elgiriyewithana, 2023). It contains more than 550,000 observations.No restricted information was used for this study.<br><b>Data Analytics Tools and Techniques</b>: The data analysis technique used is the Random Forest Machine Learning algorithm.<br><b>Justification of Tools/Techniques:</b>Random Forest is the chosen technique for this study. The algorithm “builds multiple decision trees using random samples of the data. Each tree is trained on a different subset of the data which makes each tree unique” (Geeks for Geeks, 2025). Then, each decision tree makes a prediction based on the training data, and for classification problems the final prediction is decided on majority vote, or the category that is most predicted.Random Forest is an appropriate technique for this study because the “algorithm ranks <b>features based on their importance in making predictions</b> offering valuable insights for feature selection and interpretability” (Geeks for Geeks, 2025). This element is vital to answer the research question.The algorithm works well on classification problems, as is the nature of this study, and with high volumes of data.<br><b>Project Outcomes</b>:The main project outcome is to find the most important features that characterize fraudulent transactions using feature importances from Random Forests.The second outcome is to predict whether transactions are fraudulent or not.The project provides a graph of feature importances and a report with recommendations for the company.<br><b>Sources:</b>Geeks for Geeks. (2025). <i>Random Forest Algorithm in Machine Learning.</i>https://www.geeksforgeeks.org/random-forest-algorithm-in-machine-learning/<br>Elgiriyewithana, Nidula. (2023). <i>Credit Card Fraud Detection Dataset 2023.</i>https://www.kaggle.com/datasets/nelgiriyewithana/credit-card-fraud-detection-dataset-2023/data