Assignment CE802 Machine Learning:

Design and Application of a Machine Learning System for a

Practical Problem.

Report on the Investigation:

1.) This study aims to identify and implement a suitable machine learning technique appropriate for
a problem regarding the prediction of a new hotel of a large hotel chain if opened in a given location to be
profitable or not.

Location is one of the most important factors affecting the business of a hotel. The location of the
hotel determines whether the business of the hotel would seek profit of not. The location along with the
geographical data and socio-economic data such as health, crime, population, availability to public
transport etc. would be some of the most important factors required for us to proceed with the machine
learning process. Luckily, we are provided with the historical data of successful and unsuccessful hotels
opened under the chain’s brand on similar locational conditions and also provided with the geographical
and socio-economic data about the locations and neighborhoods.

To proceed with the prediction, we must perform predictive task on the given set of data. There are
various predictive tasks available for designing a machine learning system. Some of them are,
• Classification
• Regression
• Clustering
• Rules mining etc.

We will be proceeding with the Classification type of prediction technique because classification
predicts the belonging to a class. Classification technique is preferred when the results of the model need
to return the belongingness of data points in a dataset to specific classes. In our case, we want to find out
whether the hotel would turn out to be as either class, profit which is true or class loss which is false
based on the given set of data.
Fig 1.1: The values of the coloumns from F1 to F20.
Here we can only plot the values of the coloumns from F1 to F20 because before preprocessinh, F21 and
class has missing values.

After selecting the type of technique, we are going to proceed with the preprocessing process. The
preprocessing process helps to improve the accuracy of the model. There are various preprocessing
methods, and we must select the ones that we need based on our dataset. In, our case we have over five
hundred missing values in the column F21. To fill the missing values, we can find either one of these
mean, median and mode and fill up the spaces of the missing values. We are using mean of the coloumn
to fill up the missing values.
 The next step is we have to pick a suitable classifier for our data. There are various types of
classifiers available to perform the classification process. Some of them are,
• Decision Tree classifiers
• K-Nearest neighbor
• K-means
• Random Forest
• Support vector machine
• Naïve Bayes etc.

We will be implementing some of these classifiers to see which one of those classifiers predicts the
most accurate result for our dataset. We’ll be using Decision tree mode, Support vector machine model
and Naïve Bias model and find which one of those classifiers predicts the more accurate data.

First we will use the Decision tree model to predict the outcomes. A Decision tree gets its name as
it uses a tree like model to make decisions and the decision’s possible consequences. First we split the
P2_Data csv into train and test data. We’ll train the model using the train data and use the test data to
predict the output. For the Decision tree algorithm, we get a prediction accuracy score of 81.2%.

Then we will use the Support Vector Machine model to predict the outcomes. It uses
classification algorithms for the process of group classification. In our case, the group to be true or false.
First we split the P2_Data csv into train and test data. We’ll train the model using the train data and use
the test data to predict the output. For the SVM algorithm, we get a prediction accuracy score of 64.4%.

And then we will use the Naive Bayes model to predict the outcomes. It uses Naïve Bayes’s
theorem to get the predictions. First we split the P2_Data csv into train and test data. We’ll train the
model using the train data and use the test data to predict the output. For the Naïve Bias algorithm, we get
a prediction accuracy score of 65.6%.

Out of these algorithms, Decision tree algorithms seem to predict a more accurate result compared
to the other two models. So, we’ll use decision tree algorithm for our dataset. Now, we’ll implement the
decision tree on P2_test_data csv file to predict the class colomn. After the prediction, we’ll print out the
predictions into another excel under the name, P2_test_predictions.csv which would be our desired
output.
Fig 1.2: The values of input coloumns plotted with respect to class after prediction.
 Fig 1.3: The values of input coloumns plotted with respect to class after prediction.

After prediction the values of the prediction data are plotted as a graph for the purpose of understanding.
2.) Now we’ll have to build a similar Machine learning system for a different company. But in this case
the company provided me with the data of each of the company’s business along with the numerical data
of the company’s profit or loss margin. Here the company wants us to build an algorithm to check if the
new business would be profitable along with it’s annual profit.

Fig 2.1: Plotting all the input coloumns in the form of Histogram
 Since we have to do predict the values of annual profit and the term annual profit is a continuous
outcome, we’ll use Regression technique to perform the prediction function. There are various types of
regression algorithms available, some of them are:
• Linear regression algorithm
• Elastic net algorithm
• Lasso regression algorithm
• Ridge regression algorithm.
• Logistic regression algorithm
• Decision tree regression algorithm etc.

Before selecting the type of technique, we are going to proceed with the preprocessing process. The
preprocessing process helps to improve the accuracy of the model. There are various preprocessing
methods, and we must select the ones that we need based on our dataset. In, our case we have Categorical
values in coloumns F6 and F9 we have to change the datatype from String to Float.

The next step in pre processing we’ll proceed with correlating the input coloumns with the Target
coloumn and drop the coloumns that are least affecting the target in order to improve the accuracy.

Now, that the preprocessing is completed, we’ll proceed with the regression process. First we will
use the Linear regression model to predict the outcomes. It describes the relation between dependent
variable which is Target and one or more input variables. First we split the P3_Data csv into train and test
data. We’ll train the model using the train data and use the test data to predict the output.

Then we will use the Decision tree model to predict the outcomes. A Decision tree gets its name as
it uses a tree like model to make decisions and the decision’s possible consequences. First we split the
P3_Data csv into train and test data. We’ll train the model using the train data and use the test data to
predict the output.

And then we will use the Elastic Net model to predict the outcomes.It is used in fitting the linear
and logistic models of regression. First we split the P3_Data csv into train and test data. We’ll train the
model using the train data and use the test data to predict the output.

We’ll get the accuracy and root mean square values of each model to determine which model is more
accurate to proceed with.

Model Type Accuracy Score Root mean square error

Linear regression 68.8% 689
Elastic Net 56.3% 810
Decision tree 23.5% 1071
Fig 2.2: Table containing the values of Accuracy and Root mean square error for their respective
models.

From Fig 2.2, we see that linear regression model yields more accuracy compared to the rest so we
proceed to perform linear regression to predict the Target coloumn of P3_test data. After the prediction,
we’ll print out the predictions into another excel under the name, P3_test_predictions.csv which would be
our desired output. The company can use this model for their future prediction of Annual incomes.