Check the python notebook for the complete solution Click Here
As part of an Audit team of a large company, you are tasked with identifying any irregularities in the company’s general ledger. These irregularities could be anything from an accidental typo (i.e an extra zero) to a calculated fraud. You are assisting the audit team to apply the best of Data Science to form risk-based hypotheses.
The general ledger houses a company’s accounts and many journal entries are posted to these accounts to either credit or debit balances. When checking journal entries for outliers, typically a rule-based analysis is performed, such as identifying manually entered journals that:
- Do not net to zero
- Are above a particular amount
- Are posted by people that post relatively few journals
- Are round amounts ($32,000)
- Have particular words in the narration (note: the descriptive columns were omitted to preserve the anonymity of the data)
- Are posted by certain people (i.e the CEO)
Your task is to use any analytical technique of your choice to identify outliers or anomalies in the dataset. It does not have to follow the typical rule-based approach above, but we are most interested in identifying journal postings that do not look like business as usual.
-
Given the quality of the dataset provided and the amount of information available, the best approach to detecting anomalous transactions is to use a rule-based analysis.
-
From the provided Data Dictionary, It states that the
BAT_NAMEcolumn is an ID for a single journal posting, and all the transactions under a particular BAT_NAME ID must net to zero. -
Concretely, If the journal entries of a BAT_NAME ID is not balanced (i.e, does not net to zero), we can see that journal posting(entry) is anomalous or
not business as usual -
Hence, our approach is to sum all the
Amountcolumn entries that belongs to a particularBAT_NAMEID and extract all the Journal Posting (Transactions) whooseAmountdoes not net to zero