Objective
– Analyze transactions in a general data table, categorizing anomalies into low, medium, and high risk.
– Dataset includes various transactions within the timeframe (DD/MM/YYYY to DD/MM/YYYY).
– Apply both supervised and unsupervised machine learning (ML) commands using ACL Analytics.
Scope
– General data table encompasses all transactions.
– ML techniques applied to understand patterns within the specified period.
Methodology
1. Obtain a basic understanding of business processes.
2. Import data into ACL Analytics, ensuring data integrity.
3. Apply unsupervised ML using the CLUSTER command on combined transactions.
4. For supervised ML, use the TRAIN command on 2022’s transactions to create a model.
5. Apply the model to 2023’s transactions using the PREDICT command.
Output and Results
Un-Supervised Learning:
– Use CLUSTER command on ‘AMOUNT_VALU’ field to group transactions into three clusters.
– Cluster 2 indicates two significantly different transactions, suggesting potential high-risk or anomalous activities.
– Cluster 0 transactions categorized as low-risk. – Cluster 1 transactions considered medium-risk.
Supervised Learning:
– Employ TRAIN and PREDICT commands for supervised ML.
– Predictive model, trained on randomly selected transactions, highlights three instances in 2023 where actual ‘Entered_Credit’ amount exceeds predicted value.
Conclusion:
– ACL Analytics and machine learning provide a comprehensive analysis, categorizing transactions into high, medium, and low risk.
– Software’s control points enhance efficiency, alerting to anomalies and guiding focus at the transaction level.
– Machine learning proves invaluable in improving the accuracy and effectiveness of transaction risk assessment.
