I am working on a project where I have hundreds of Excel files containing financial transactions (e.g., stocks, mutual funds, etc.) in varying formats. Each file can contain one or more tables, and the tables are located at different positions across different sheets with respect to different brokers like zerodha, groww, etc.

My goal is to train a machine learning model that can automatically identify and extract the relevant table(s) from any given Excel file, even if the format is unfamiliar.

** This is the data :**

https://drive.google.com/drive/folders/1YixwjLg2ZskRXMI5WMjns5kD1Ujfpphd?usp=sharing

Every broker has a different format in which they give this data to user so I cannot parse it manually as format changes frequently.

Like in image, pdf files we train on a no. of files and then new format is given, the model can give the required data with precision and accuracy even if the format is changed. The challenge is to perform this on excel files.

I tried annotating the excel files with the coordinates of location of table and type of data in each cell within that range but found them to be irrelevant as a feature.
Aside from machine learning, I tried using fuzzy-wuzzy library so that any time a column header is changed by broker, it maps to a single header – E.g.- ‘buy date’, ‘entry date’ will map to ‘purchase date’ but that doesn’t solve the problem.
Please help me, I have searched and researched extensively but can’t figure it out