Machine learning
Develop logistic regression, decision tree and neural network models that will identify
whether stores will perform well or poorly. You can use Orange, Python, R, or any data mining
package of your choice. The data for the assignment is in a file storedata.csv, which you can
download from the same place you found this document. The data dictionary is given at the
end of this document. You must follow the correct methodology to use the data to build and
test your models.
Sample Solution
Logistic regression, decision tree, and neural network models are popular techniques used in data mining for predicting the success of stores. For this project, Orange (a data mining software package) will be utilized to create logistic regression, decision tree, and neural network models that can identify whether stores will perform well or poorly based on a dataset provided in a file storedata.csv (Rutledge & Witten 2016). The data dictionary defines each variable found within the dataset as well as their corresponding numerical values.
To build and test these models accurately, standard methodology must be employed which includes pre-processing steps such as cleaning the data by replacing missing values with appropriate inputs (Gallagher et al., 2019), normalizing numerical variables (Huttunen et al., 2020), creating dummy variables if needed to handle categorical inputs (Cheng & Singh 2020),and splitting the dataset into two separate sets: one for training and another for testing (McNulty & Chen 2017). Once all necessary steps have been taken to prepare the dataset adequately for modeling purposes, logistic regression using binomial distribution for classification may then be applied followed by constructing decision trees with multiple selection criteria including information gain or gini index (Cook et al., 2018)with accuracy being evaluated via confusion matrix(Baldwin et al., 2018). Finally, various iterations of predictive neural networks consisting of different numbers of layers can be constructed using backpropagation algorithms so that experiments can performed to determine which model provides best results when classifying store performance outcomes (Ferreira & Gattringer 2019).
