- Creating a model which learns to forecast future transactions for your account.
- Exploratory Data Analysis (EDA) is done to understand and gain insights from the data.
- Previous bank statements are used to collect the data.
- Refining and cleaning of data takes place for better use.
- Time series model is fitted on the data for forecasting.
Python Version: 3.10.9
Jupyter Lab Version: 3.5.3
Packages: pandas, numpy, matplotlib, seaborn, pmdarima, statsmodels, scipy
Project Format: https://www.youtube.com/playlist?list=PL2zq7klxX5ASFejJj80ob9ZAnBHdz5O1t
SARIMAX Article: https://towardsdatascience.com/time-series-forecasting-with-arima-sarima-and-sarimax-ee61099e78f6
Markdown Cheatsheet: https://github.com/adam-p/markdown-here/wiki/Markdown-Cheatsheet
- Bank statements are used to collect data of transactions.
- It consists of Date Of Transaction, Transaction Amount, Type of Transaction, Deposit/ Withdrawl.
- Data collection is done for as long duration as possible for better results.
- All the data is imported in an excel file for further processing using pandas in python.
After collection of data, I needed to clean to remove unwanted values and anything that will hinder its perfomance
- Data collected is clean for better analysis and understanding.
- Removing of null values, creating new columns for UPI transactions.
- Converting multiple transactions on a day to single transaction for further processing.
- Calculating monthly transactions and importing a new dataframe based on it.
Exploratory Data Analysis is done on both Daily, Monthly Transactions
Kernel Density Estimate Of Monthly Withdrawls
- Analysing all the transactions with respect to date
- Using two models to gain best results
- ARIMA model without seasonality
- SARIMAX model with seasonality to learn trends
Arima model was applied, but its results were static.
Errors of the model were as follows:
- Mean Absolute Error: 3282.0327896414537
- Mean Absolute Percentage Error: 3.022257511793852e+18
- Root Mean Squared Error: 5817.4732058623495
SARIMAX model was applied for better result and understanding seasonality:
Errors of the model were as follows:
- Mean Absolute Error: 3971.822417234379
- Mean Absolute Percentage Error: 4.645845057110243e+18
- Root Mean Squared Error: 5697.147329455461
