Classification of Dry Beans

For the purposes of this assignment you will develop two different classification predictive models to classify

different types of dry beans. You have to write a report wherein you provide responses in clear narrative on the

aspects enumerated below, under appropriate section headings. Note that code will not be evaluated. Tables

and figures will also not be considered if these tables and figures are not accompanied by your own explanation

of what these tables and figures portray.

Complete the assignment in the following steps:

Download the DryBeanDataSet$874.$xlsx dataset. The dataset contains $13611$ instances, $20$ descriptive

features, and the class feature Class in column U$.$

Without changing anything in the provided dataset, provide an analytics base table wherein you charac$-$

terize all of the features of the dataset.

You now have to very carefully explore the dataset to identify data quality issues. For this part of

your report, only identify the data quality issues and provide justifications for these issues. One of the

data quality issues is that some of the class labels are missing. Excude this data quality issue from the

discussion.

Based on your analysis above, decide on two different machine learning approaches that you will employ

to construct a predictive model for this problem. Give justifications for why you have selected these two

approaches for this problem.

For this part of the assignment, only focus on the data quality issues with respect to the descriptive

features. For each of the machine learning approaches, discuss the data$-$preprocessing steps that you have

implemented to optimally transform the dataset for that specific machine learning approach and to correct

data quality issues. Note: do not do unnecessary data transformations. Carefully think about the data

transformations needed for your selected machine learning algorithms. Provide justifications for each of

these pre$-$processing steps. Should you decide not to address a data quality issue, justify this decision.

When you pre$-$process the dataset, make sure that you do not change the order of the instances in the

dataset.

For this part of the assignment, only use the instances that have a known class label. Develop the two

predictive models and evaluate the performance of the two models. Make sure to construct optimal

configurations of your chosen models both with respect to architecture and values for control parameters.

Describe the process that you have followed to produce an optimal configuration for each model. For this

purpose, carefully decide on the performance metrics that you will use. Conclude on which one of the two

approaches is best for this problem, and support your conclusion with justifications. For the purposes of

this assignment, make sure to report the performance based on a $k-$fold cross$-$validation. Decide on the

number of folds with a justification.

For the last part of the assignment, focus returns to those instances that have a missing class label. Make

use of $k-$nearest neighbour to impute a class label for each of these instances. Describe how you have

used $k-$nearest neighbours for this purpose. You have to decide on the value of $k$ with justification. In a

table list the instance number and the imputed class label. Then, for your best model identified above,

retrain the model on the new datasets with the imputed class labels. Report on the performance of the

model, compared to the results obtained from step $6$ above and conclude on the efficacy of the $k-$nearest

neighbour self$-$labeling process.