ATM Cash Prediction using Machine Learning approach

Project Title

ATM Cash Prediction using Machine Learning approach

Project Area of Specialization Artificial IntelligenceProject Summary

The advent of Automatic Teller Machines (ATMs) enable self-service, time-independent, easy to use, mechanism through which a financial institution supports large number of cash delivery transactions. Although, ATMs are now growingly used for non-cash transactions, but still cash transactions are the main essences of these machines. The rapid adaptation and standardization of these network give rises to many challenging problems that requires intelligent management of these resources. One of the most challenging problem is cash demand forecasting for individual ATM on daily-basis. This is to predict exactly how much amount of cash is required to serve the requested users per day. The problem has two sides of consideration for optimization (i) if a large amount of cash lying idle at the ATM, it would cost an interest opportunity loss for institution, and (ii) if the machine is empty it would cost the customer dissatisfaction. Considering on average an institution has hundreds of ATMs on their network, it is quite critical to decide about the exact amount required daily for each individual machine. This final year project is an attempt to predict the daily cash requirements for each individual machine, by utilizing the low-level transactional data, customer’s profiles, ATM meta-data and time. A Deep Neural Network (DNN) model is proposed for forecasting daily cash requirements of each ATM. The proposed approach will be implemented and evaluated using Mean Absolute Percentage Error (MAPE) across all the ATMs. 

Project Objectives

Given the data generated at an ATM for a period of ’T’ days we want to predict the amount required at some specific ’d’ day for upcoming
transactions.

Project Implementation Method

We propose to predict the amount required for the upcoming transactions based on the past data using Machine Learning models. The first phase will be data handling, in which the goal is to find the NaN and missing values and replace them with  values produced using interpolation which can effect the results. The next phase will be the division of the data for the train and test purpose. After that ML models will be trained on the 70percent of the past data and the rest of the data (30 percent) will be used to test our results.

Following are the list of the models on which the training and testing of the data will be done:

Neural Networks

Support Vector Machine

Regression Model

After that Deep Learning applied on the approach

Benefits of the Project

Automatic Teller Machines (ATMs) are the devices financed and managed by financial institutions to provide facility to humans to
make transaction in public places. Banks usually refill ATMs per day and this cash replenishment is manual till now, cash in ATMs is kept upto 40 percent more than its real need, This excess amount leads to business loss as that amount can be used by Banks for investments and various purposes that can generate revenue for them therefore we propose automatic forecasting mechanism to make cash
replenishment prediction more accurate to avoid business loss and
customer dissatisfaction. To fulfill this task we are using real data of 20 ATMs from 8 months and to make prediction much accurate we will take into account seasonality effects as well as location effect, we will use historical data of ATM to predict future demand through Machine learning approach.

Technical Details of Final Deliverable

The final delivarables will be a web based platform with role based access where the model will be deployed. So the concerned authorities can access the website and upload the data and get the result.

Final Deliverable of the Project Software SystemCore Industry ITOther Industries Finance Core Technology Artificial Intelligence(AI)Other Technologies Artificial Intelligence(AI)Sustainable Development Goals Sustainable Cities and Communities, Responsible Consumption and ProductionRequired Resources

Item Name	Type	No. of Units	Per Unit Cost (in Rs)	Total (in Rs)
			Total in (Rs)	70000
GTX 1080	Equipment	1	70000	70000