Placement Prediction using Various Machine

Learning Models and their Efficiency

Comparison

Abstract:-The purpose of a placement predictor is to determine a student's likelihood of being placed in a

company based on the company's requirements. Numerous criteria are taken into account by the placement
predictor, which can be used to evaluate a student's competence level. Certain criteria are derived via
university-level assessments, while others are acquired from tests administered within the placement
management system. By combining these data points, the predictor may determine with accuracy whether or
not the student will be hired by a company. Data from former pupils are used for training the predictor.

The challenge, however, was to identify an appropriate classification algorithm that could complete the
task for our data set with the highest level of accuracy. The accuracy of an algorithm varies based on the
kind of problem it has to answer and the data set it uses. Thus, we made the decision to assess the
accuracy levels of four algorithms—KNN, SVM, Random Forest, and Logistic Regression—in relation to our
problem and data set. We would be able to choose the best algorithm to use our predictor in the placement
management system.
In order to determine the algorithms' accuracy, we used the data set we had obtained to train each
algorithm and compared the results to some test data. We can quickly determine the True Positive, True
Negative, False Positive, and False Negative for each algorithm. It was just a matter of utilizing the accuracy
equation.

We aim to develop a placement predictor as a part of making a placement management system at college level which
predicts the probability of students getting placed and helps in uplifting their skills before the recruitment process
starts. We are using machine learning for the placement prediction. We consider K-nearest neighbour

Regression, Random Forest to classify students into appropriate clusters and the result would help them in
improving their profile.And accuracy of respected algorithms are noted and With the comparison of various
machine learning techniques, this would help both recruiters as well as students during placements and related
activities.
A. Key words
 Classification
 Random Forest
 Regression
 Dimensional Space
 Optimization

B. Prediction system
In this paper we use machine learning techniques to predict the placement status of students based on a dataset. The
parameters in the dataset which are considered for the prediction are Quantitative scores, LogicalReasoning scores,
Verbal scores, Programming scores, CGPA, No. of hackathons attended, No. of certifications and current backlogs
number. The placement prediction is done by machine learning using Logical Regression, Random Forest, KNN,
SVM.
Table 1:- Dataset used for Prediction and Analyss
C. Architecture Diagram

Fig 1:- Architecture for Data Processing , Model Training, Prediction and Accuracy check.

The data frame for the machine learning algorithm is created using pandas library based on the above sample dataset.
The handling of null data fields is carried out by dataset.fillna(method='ffill'). We use sklearn which is an

efficient tool for predictive analysis. And we import

train_test_split in sklearn for creating training and test sets from the dataset. Standardisation is done by standardscaler in
sklearn.preprocessing. Based on the respective algorithm it predicts the placement of each student and accuracy can be viewed
from the confusion matrix.
II..METHODOLOGY
Machine Learning can be done using many available algorithms. Each algorithm has its own set of merits and demerits, these can chan
for the type of dataset being used or for the type of problem we have at hand. Given below are some of the algorithms that we used.
detailed description accompanies each algorithm.

A. KNN
KNN stands for k-nearest neighbors. This is a simple
algorithm that can be used to solve classification and regression type problems. It is a supervised machine learning
algorithm, meaning labels are used.

The basic working of this algorithm revolves around the concept that similar things are always in close proximity
within each other. So, for this algorithm to provide any fruitful results, this is an assumption that is taken. Similarity in
KNN is expressed using distance,
The algorithm:
 Load the data to be used. .this model.

 Initialize the value of K to a chosen number of neighbors

 For each entry in the data:

The distance between the current example and query example is to be calculated from the data. The index of the example
and its distance is added to an ordered collection.

 Sort this ordered collection in ascending order by the distances

 Pick the first K entries from this sorted collection

 The labels of the selected K entries are taken.

 Now, if it is regression, calculate and return the mean of the K labels. But if it is classification, calculate and return
the mode of the K labels

This is a simple algorithm to implement KNN.

However the error of the algorithm depends on the value selected for K. So, to find the K that is best suited for the data
given, it is advised to run the algorithm many times with different values of K, so that the K with the least error can be found.

 Advantages:
 This is a fairly simple and easy-to-implement algorithm
 Building a model, tuning several parameters or making additional assumptions are not required
 This is a versatile algorithm, being able to be used in regression, classification and even search problems.

 Disadvantages:
 The algorithm becomes significantly slower as the number of examples and/or predictors/independent
variables increases.

The disadvantage that KNN provides makes it an impractical choice for use where predictions need to be made more
rapidly. However, provided that one has enough computational power at their disposal, then KNN can be used in problems where
similar items have to be identified.
 B. SVM
supervised machine learning algorithm that can be used for both classification and regression problems. However, it is mostly used
for classification problems.

A point in the n-dimensional space is a data item where the value of each feature is the value of a particular coordinate.
Here, n is the number of features you have. After plotting the data item, we perform classification by finding the hyper-plane
that differentiates the two classes very well.

Now the problem lies in finding which hyper-plane to be chosen such that it is the right one.

Scikit-learn is a library in Python which can be used to implement various machine learning algorithms and SVM too can
be used using the scikit-learn library.

 Advantages:
 This algorithm performs best when there is a clear
margin of separation

 Effective in high dimensional spaces

 If the number of dimensions is greater than the number
of samples, the algorithm would be able to perform
better
 It is memory efficient

 Disadvantages:
 Performance is affected when large data sets are used as
the required training time is more.
 Performance is also affected when the data set has too
much noise
 SVM doesn’t directly provide probability estimates,
rather a computationally intensive five-fold cross-
validation is required.
 C. Logistic Regression
Logistic regression is a classification technique

it s very good for binary classification. It's decision boundary which is generally linear derived based on probability
interpretation. The results are in a nonlinear optimization problem for parameter estimation. Parameters can be
estimated by maximising the expression using any nonlinear optimization solver.
The goal of this technique is given a new data point, and predict the class from which the data point is likely to have
originated. Input features can be quantitative or qualitative.

Instead of a hyperplane or straight line, the logistic regression uses the logistic function to obtain the output of a
linear equation between 0 and 1.

The function is defined as logistic(x)=1/(1+exp(-x))

Fig 2:- Logistic regression

 Advantages

 Logistic Regression is good for linearly separable

 It is efficient to train and easy to interpret and implement.
 It not only gives a measure of how relevant a predictor is, but also its direction of association.

 Disadvantages
 It is useful only for predicting discrete functions.
 It should not be used If the No. of observations in the
dataset are lesser than the number of features.
 Assumption of linearity between the independent and
dependent variables.
D. Random Forest
We have a plethora of classification algorithms at our disposal, including, but not limited to, SVM, Logistic

regression, decision trees and Naive Bayes classifier, just to name a few. But, in the hierarchy of classifiers, the Random Forest
Classifier sits near the top. The random forest classifier is a group of individual decision trees and so, we shall look into
how decision trees work.

It is basically a flowchart-like structure in which each node excluding the leaf node is a test on a feature (i.e, what will be
the outcome if some activity, such as flipping a coin, is done), leaf nodes are used to represent the class label (the decision
taken after all features are computed) and branches represent the conjunctions of features that lead to those class labels.
The classification rules of a decision tree are the paths from the root node to the leaf node.

So then, now let us look into random forest classifiers. As mentioned earlier, it is a collection of decision trees. The basic
idea behind random forest is “the wisdom of the crowds”. It is a powerful concept wherein a large number of uncorrelated
models, or in this case trees, operating as a group, would provide a much more solid output than any of the constituent models.

So, in a random forest, each individual tree with different properties and classification rules would try to find an
appropriate class label for the problem. Each tree would give out its own answer. A voting is done within the random forest to
see which class label received the most votes. The class label with the most votes would be considered the final class
label for the problem. This provides a more accurate model for class label prediction.

 Advantages:
 It can balance errors in data sets where classes are imbalanced
 Large data sets with higher dimensionality can be handled
 It can handle thousands of input variables and could identify the most significant variables and as such, it is
a good dimensionality reduction method

 Disadvantages:
 It does more good of a job for classification
problems rather than regression problems
as it finds it harder to produce
continuous values rather than discrete ones
 IV. RESULT AND ANALYSIS

The final result of performing various machine algorithms are mentioned in the table below. We considered KNN,
Logistic Regression, Random Forest and SVM for the analysis. We trained and predicted the placement status of
students based on the same dataset and found the True Positive, False Positive, False Negative, True Negative and
accuracy of each algorithm. And it is tabulated in the table below.

Table 2:- Accuracy of algorithms

Data Sample

S.No Cgpa Iq placement

1 6.8 123 1
2 5.9 106 0
3 5.3 121 0
4 7.4 132 1
5 5.8 142 0
6 7.1 48 1
7 5.7 143 0
8 5 63 0
9 6.1 156 1

Table 1:- Dataset used for Prediction and Analysis

 V. CONCLUSION
Placement prediction system is a system which predicts the placement status of final year B-Tech students. For data
analysis and prediction different machine learning algorithms are used in the python environment. We analyse the accuracy of
different algorithms and it is shown in the above table. It is clear that SVM gives an accuracy of 100. Logistic Regression is also
good which gives an accuracy of 97.59 based on the given dataset. The accuracy of Machine learning algorithms may differ
according to the dataset. From the result from our analysis it is clear that SVM, Logistic Regression, Random Forest, KNN are
good for binary classification problems since they all give accuracy of above 95. Some recruiters consider GATE scores
and history of backlogs which we didn’t include in our dataset. In such rare cases these results may change.

REFERENCES

[1]. Shreyas Harinath, Aksha Prasad, Suma H and Suraksha A. Student Placement Prediction using Machine
Learning, International Research Journal of Engineering and Technology (IRJIET) Volume : 06 Issue: 04 April 2019
Senthil Kumar Thangavel, Divya Bharathi P and
[2]. Abhijith Shankar. Student Placement Analyzer: A recommendation System Using Machine Learning, International
Conference on advanced computing and Communication systems (ICACCS-2017), Jan 06- 07,2017, Coimbatore, INDIA.
K. Sreenivasa Rao, N. Swapna, P. Praveen Kumar
Educational data mining for student placement prediction using machine Learning algorithms Research Paper,
[3]. International Research Journal of Engineering and Technology (IRJIET) 2018