QBUS2820 Assignment 1 (30 marks)
March 20, 2024
1 Background
Developing a predictive model for ATM cash demand is an important task for every bank. Sup-pose that you are employed by a bank, and your task is to optimise the bank’s cash management by making smarter decisions about reloading its ATM network.
The variable Withdraw in the dataset ATM_training.csv is the total cash amount withdrawn per day from an ATM, recorded from the ATM network of a bank. The response variable and covariate variables are described in the following table.
Variable Description
Withdraw The total cash withdrawn a day (in 1000 local currency)
Shops Number of shops/restaurants within a walkable distance (in 100)
ATMs Number of other ATMs within a walkable distance
Downtown =1 if the ATM is in downtown, 0 if not
Weekday = 1 if the day is weekday, 0 if not
Center =1 if the ATM is located in a center (shopping, airport, etc), 0 if not
High =1 if the ATM has a high cash demand in the last month, 0 if not
Your task is to develop a linear regression model for predicting the cash demand Withdraw based on the covariates.
You are also given the dataset ATM_test_without_Withdraw.csv, which is the real test dataset ATM_test.csv with the Withdraw column deleted. The test dataset ATM_test.csv (not provided) has the same structure as the training data ATM_training.csv.
1.1 Test error
For the measure of prediction accuracy, please use mean squared error (MSE), computed on the test data. Let ybi be the prediction of yi where yi
is the i-th withdraw in the test data. The test error is computed as follows
where ntest is the number of observations in the test data.
2 Submission Instructions
1. Please submit THREE files (or more if necessary) via the Canvas site:
• A document file, named SID_Assignment1_document.pdf, that reports your data analysis procedure and results. You should replace “SID” in the file name with your student ID.
• A Python file, named SID_Assignment1_implementation.ipynb that implements your data analysis procedure and produces the test error. You might submit additional files that are needed for your implementation, the names of these files must follow the same format SID_Assignment1_<name>.
• A csv file SID_Assignment1_Withdraw_prediction.csv that contains the prediction of the withdrawals for the ATMs in the dataset ATM_test_without_Withdraw.csv.
2. About your document file SID_Assignment1_document.pdf
• Describe your data analysis procedure in detail: how the Exploratory Data Analysis (EDA) step is done, what and why the methods/predictors are used, etc. with suffi-cient justifications. The description should be detailed enough so that other data scien-tists, who are supposed to have background in your field, understand and are able to implement the task. All the numerical results are reported up to four decimal places.
• Clearly and appropriately present any relevant graphs and tables.
• The page limit is 15 pages including EVERYTHING: appendix, computer output, graphs, tables, etc.
3. The Python file is written using Jupyter Notebook, with the assumption that all the necessary data files (ATM_training.csv and ATM_test.csv) are in the same folder as the Python file.
• The Python file SID_Assignment1_implementation.ipynb must include the following code
import pandas as pd
ATM_test = pd .read_csv( 'ATM_test.csv')
# YOUR CODE HERE: code that produces the test error test_error
print(test_error)
The idea is that, when the marker runs SID_Assignment1_implementation.ipynb, with the test data ATM_test.csv in the same folder as the Python file, he/she expects to see the same test error as you would if you were provided with the test data. The file should contain sufficient explanations so that the marker knows how to run your code.
• You should ONLY use the methods covered in the lectures and tutorials in this assign-ment. You are free to use any Python libraries to implement your models as long as these libraries are publicly available on the web.
3 Marking Criteria
This assignment weighs 30 marks in total. The content in SID_Assignment1_document.pdf con-tributes 15 marks, and the Python implementation contributes 15 marks. The marking is struc-tured as follows.
1. The accuracy of your forecast: Your test error will be compared against the small-est test error (among all students including the teaching team). The marker first runs SID_Assignment1_implementation.ipynb
• Given that this file runs smoothly and a test error is produced, the 15 marks will be allocated based on your prediction accuracy, compared to the smallest MSE, and the appropriateness of your implementation.
• If the marker cannot get SID_Assignment1_implementation.ipynb run or a test error isn’t produced, some partial marks (maximum 5) will be allocated based on the appro-priateness of SID_Assignment1_implementation.ipynb.
2. Your report described in SID_Assignment1_document.pdf: The maximum 15 marks are allocated based on
• the appropriateness of the chosen forecasting method.
• the details, discussion and explanation of your data analysis procedure.
See the Marking Criteria for more details.
4 Errors
If you believe there are errors with this assignment please contact the teaching team.