Reference no: EM133672508

Predictive Analytics

Assignment: Building and Evaluating Predictive Models

Objective:
Revise BUS5PA material on predictive modelling
Demonstrate knowledge of data exploration and selection of variables to apply for the predictive models
Demonstrate knowledge of building different types of predictive models using R
Demonstrate knowledge on comparing and evaluating different predictive models
Relate theoretical knowledge of predictive models and best practices to application scenarios

Business Case

A used car online selling company in the USA is in the process of updating their car price assessment method where they want to apply a data driven technique. The trial dataset consists of 25 variables describing 20673 car sales from 2019 to 2020. The management is very keen to apply predictive modelling for this task where the trail data set is to be used to build and evaluate predictive models to ascertain the feasibility of such an approach. The company has outsourced the task to you.

Part A - Problem Formulation

The objective of this section (Part A) is to introduce students to the ‘domain understanding and familiarisation' phase data analysts go through prior to the actual analytics. Since you may have to carry out analytics projects in different domains in the future, where you may not have domain knowledge, it is important to develop this skill.

Carryout an exploratory study to identify the background and relevant aspects of used cars which influence their value (you can research USA and Australia) and methods used for price evaluation and assessment of used cars?
Identify the data sources that would contain information useful for value assessment of used cars. What is the possible format of such information? Will you face any problems in accessing these data?
What variables would be useful to build a predictive model to assess the used cars?

You are expected to prepare a brief report with the answers to the above questions (Maximum 3 pages)

Part B - Data Exploration and Cleaning

Use the provided dataset to answer this section. You are given access to 24 variables that are directly related to used car sales from the above-mentioned dataset. Most of these variables are similar to the type of information that an assessor will use to evaluate and assess the price of a used car (e.g., when was it made? What is the length and width of the car? What type of wheel system? Number of seats?). You need to answer the following questions with evidence and justifications.

a. Which variables are numerical and categorical? Which are ordinal, Nominal, discrete and continuous? Prepare a table similar to below given table

b. What are the methods for transforming categorical variables?
c. Carry out and demonstrate data transformation where necessary. (Show your R script lines related to this in the report as well)

a. Calculate following summary statistics:
mean, median, max, min and standard deviation for each of the numerical variables,
count for each categorical variable.
b. Is there any evidence of extreme values? Provide the relevant diagrams to identify the extreme points and briefly discuss.

Plot histograms for each of the continuous variables. Based on the histogram and summary statistics that you got in 2(a), answer the following and provide brief explanations:
Which variables have the largest variability? Justify the reason.
Which variables seem skewed? Justify the reason.
Are there any values that seem extreme? Justify the reason.

a. Which, if any, of the variables have missing values?
b. What are the methods of handling missing values?
c. Apply the 3 methods of handling missing values and demonstrate the output (summary statistics and transformation plot) for each method in (4-a). (hint: the objective is to identify the impact of using each of the methods you mentioned in 4-a on the summary statistics output above). Which method of handling missing values is most suitable for this data set? Discuss briefly referring to the data set.

a. Evaluate the correlations between the variables.
b. Which variables should be used for dimension reduction and why? Carry out dimensionality reduction.
c. Explore the distribution of selected variables (from step 5-a) against the target variable. Explain.
Part C (50%) - Building predictive models

Building Linear Model
Build a regression model with the selected variables after dimension reduction (part B-5b) and evaluate the model using evaluation metrics.
Carry out further feature selection to build a better linear model.
Justify on the feature selection process. You need to try out at least another 2 linear models to identify the optimal model.
Provide the relevant evaluation metrics/diagrams/tables of each model in your report.
Compare these regression models based on evaluation metrics and provide the formula for each regression model.

Decision Tree Modelling
Build a decision tree with the selected variables after dimension reduction (part B- 5b).
Evaluate the decision tree model and carry out pruning to build a better decision tree model. You need to try out at least 3 decision trees to obtain the optimal tree.
Compare these decision tree models based on evaluation metrics and provide the tree plot for each model and explain the outputs.

Model Comparison
Why do we need to build several models in both regression and decision trees (as requested in Part C question 1 and 2)?
Compare the accuracy of the selected (optimal) linear model and (optimal) decision tree and discuss and justify the most suitable predictive model for the business case.