Development and Comparative Analysis of an Instance-Based
Machine Learning Classifier
Govind Agarwal
a
, Chirag Goel
b
, Chunduru Sri Abhijit
c
and Alok Chauhan
d
School of Computing Sciences, Vellore Institute of Technology, Chennai, Tamil Nadu, India
Keywords: Classification, Machine Learning, Lazy-Learner.
Abstract: Classification algorithms make it easy to classify many real-world problems, but they come with some cost.
The existing classification algorithms have complex architectures, which can sometimes make the
classification task tedious. This paper introduces a classification algorithm, which aims to improve upon
existing methods by incorporating class count as a target feature. In this study, we attempt to offer a
classification method that works with three different categories of datasets, viz., categorical, numerical, and
a mixture of categorical and numerical. Firstly, for each input feature attribute, proposed algorithm counts the
majority class of the target variable to train the model. Then it determines which class has appeared the most,
after computing the majority class for each input characteristic. Final output of the classification algorithm
would be the class that showed up the most. If there is a tie in the number of attributes, the class with the
greater total count wins. Instance can belong to any class if the total count is also the same. Obviously, any
attribute, which has the same count across all classes, is redundant or has no bearing on classification. This
classification process is compared against several machine learning methods like KNN, logistic classifier and
other models. Experimental results on various benchmark datasets demonstrate that the proposed algorithm
is reliable and is promising with respect to several state-of-the-art classification methods in terms of
classification accuracy as well as computational efficiency.
1 INTRODUCTION
Classification machine learning algorithms are a
subset of supervised learning techniques, designed to
discover a mapping between input data and output
labels. Many fields have developed and adapted
classification algorithms to tackle and automate a
variety of practical problems (I.H. Sarker, 2021).
Classification algorithms are defined here as a task of
identifying the correct category of unseen data, based
on the characteristics of previously seen classes
(Tammy Jiang, 2020).
There are numerous classification algorithms, each
having its advantages and disadvantages. Each
algorithm utilizes a different approach to divide the
data into classes, with some depending on simple,
linear decision boundaries and others using more
complicated, nonlinear ones. Many factors must be
a
https://orcid.org/0009-0008-4673-8090
b
https://orcid.org/0009-0009-8336-0987
c
https://orcid.org/0009-0001-4108-2888
d
https://orcid.org/0000-0002-8309-3403
considered while deciding on an algorithm, such as
the dataset's size and complexity, the type of
characteristics to be used as inputs, and the level of
precision and interpretability that must be achieved.
The type of classification developed during this
research is a lazy learner. K-nearest neighbours and
locality-sensitive hashing are two famous examples
of classification methods for lazy learners. In addition
to selecting an effective method, it is essential to pre-
process and prepare the data before training the model
to ensure that it is representative, balanced, and free
of mistakes or outliers.
This research paper's primary objective is to look into
following three variants for classification algorithm:
For datasets containing only categorical values
For datasets containing only numerical values
For datasets containing a mixture of categorical and
numerical values.
434
Agarwal, G., Goel, C., Abhijit, C. and Chauhan, A.
Development and Comparative Analysis of an Instance-Based Machine Learning Classifier.
DOI: 10.5220/0012509800003739
Paper published under CC license (CC BY-NC-ND 4.0)
In Proceedings of the 1st International Conference on Artificial Intelligence for Internet of Things: Accelerating Innovation in Industry and Consumer Electronics (AI4IoT 2023), pages 434-440
ISBN: 978-989-758-661-3
Proceedings Copyright © 2024 by SCITEPRESS – Science and Technology Publications, Lda.
Performance of above variants is compared against
following state-of-the-art classification algorithms:
Logistic regression
KNN
Decision trees
Gradient boosting
The presentation of the paper is as follows. Second
section begins with an analysis of prior research on
classification algorithms. Following this, there is
discussion about the proposed classification
methodology. Next, the time complexities of various
models are detailed. Experimental findings are
presented in the next section. Finally, the findings of
the study and recommendations for future research
work are presented.
2 RELATED WORK
An empirical analysis about the effectiveness of
supervised learning on high-dimensional data is
carried out by (Rich Caruana et al., 2008). The
authors implement several machine learning
algorithms like support vector machine (SVM),
artificial neural network (ANN) and others. These
models are evaluated on the basis of performance
metrics, namely, accuracy, root mean square error,
and area under the ROC metric curve. During the
study, 11 binary datasets of very high dimensionality
are evaluated and it is concluded that random forest
(RF), ANN, SVM and boosted trees outperform all
other models. It is also observed that the least
performing methods are naive Bayes and perceptron.
The study also indicates that boosted trees perform
well in lower dimensionality datasets but when it is
applied above 400 dimensions, it tends to over fit.
(Chongsheng Zhang et al., 2017) carry out an
empirical study on various emerging classifiers like
extreme learning machine (ELM), sparse
representation classifier (SRC) and others. These
classifiers are compared with traditional classifiers
like random forest, k- nearest neighbors (KNN) etc.
During the study, 71 datasets are experimented to
validate the effectiveness of the models. The results
indicate that the stochastic gradient boosting decision
trees perform well in supervised learning. (Jingjun Bi
et al, 2018) propose a new machine learning method
based on multi class imbalance, namely, Diversified
Error Correcting Output Codes (DECOC). To
validate the effectiveness of their model, they
perform experiments on 17 multi class imbalance
datasets. The results indicate that the DECOC achieve
best results in terms of accuracy (ACC), area under
the ROC Curve (AUC), geometric mean (G-mean)
and F- measure. (Amanpreet Singh et al., 2016)
compare various supervised machine learning
algorithms on the various datasets, on the basis of
accuracy, speed, comprehensibility and speed of
learning. The authors employ Bayesian networks,
naive Bayes, KNN, etc. The authors suggest that
choice of an appropriate algorithm depends on the
dataset and type of classification problem. The
authors conclude from the experimental results that
the tree-based algorithms perform better than the rest
of the algorithms. According to (Rich Caruana et al.,
2006), multiple performance criteria are used to
compare learning models in various domains. A
model may perform well on one measure but poorly
on another. Multiple performance measures assess
various trade-offs in prediction. As a result, they
evaluate algorithms based on a relatively wide range
of performance indicators. The authors compare the
ten supervised algorithms using eight distinct
performance metrics. They examine the performance
indicators before and after using Platt scaling and
isotonic regression to calibrate the outputs. They
come to the conclusion that calibrated boosted trees
outperform other methods in all eight measures.
Random Forest is at the second place. Logistic
regression and naive Bayes fare the worst. They also
find that calibration with either Platt scaling or
isotonic regression enhances SVM, stumps, and
Naive Bayes performance. (Henry Brighton et al.,
2002) start their study by detailing some practical
challenges in classification algorithms. The main
argument they make is that reduction methods have,
historically, been seen as generic solutions to the
issue of instance selection. Their studies of, how
various schemes function and how well they perform
in different contexts, lead them to believe that the
success of a scheme is strongly reliant on the structure
of the instance-space. They contend that one selection
criteria is insufficient to ensure excellent overall
performance. They conclude that for the vast majority
of classification issues, border instances are crucial to
class discrimination. Their algorithm competes with
the best effective current methods in 30 fields.
(Saksham Trivedi et al., 2021) use ML algorithms in
many fields of study. They come to the conclusion
that assignment structure has the greatest impact on
algorithm selection in machine learning. They assert
that SVM and neural networks are more valuable due
to their multidimensionality despite the fact that logic
systems are ordinarily capable of handling
differential/categorical characteristics. For neural
network models and SVMs to achieve maximum
accuracy, a large sample size is required, whereas NB
only requires a small amount of data. Makdah et al.,
Development and Comparative Analysis of an Instance-Based Machine Learning Classifier
435
2019 have noted that under the nominal conditions,
the models perform well. Based on the numerical
results obtained from the experiments done by the
authors it is suggested that the accuracy- sensitivity
trade-off is purely determined by the statistical
characteristics of the data and cannot be enhanced by
adjusting or enhancing the level of complexity of the
algorithms. In the paper, the authors have presented a
study about trade-off among a binary classification
algorithm's accuracy and its susceptibility to
uncontrolled modification of data and conclude that a
classification algorithm's accuracy can only be
maximized at the price of its sensitivity, given a set of
moderate technical assumptions. As a result, there is
a basic trade-off between the performances of a
classification system in conventional and adversarial
settings respectively. (Vaishali Gangwar, 2012)
present a summary of the categorization of
unbalanced data sets. In the study, it is observed that
sampling is the most often used strategy to deal with
unbalanced data and in case of locally trained
classifiers, oversampling outperform under sampling
approach but this scenario is inverted in the case of
global learning. However, the researcher demonstrate
that hybrid sampling strategies outperform
oversampling and under sampling. The research
suggests that in order to handle uneven data, solutions
based on modified support vector machines, rough
set-based minority class focused rule learning
approaches, and cost sensitive classifiers can be used
as an alternative to the classical approach. (José A.
Sáez et al., 2013) conduct a comparative analysis of
the noise robustness of single classifiers and Multiple
Classifier Systems (MCS). The authors attempt to
determine the efficacy and robustness of singular
classifiers when trained on noise datasets. It is
concluded that the robustness of the model against
noise depends on the noise level, and in the majority
of cases the MCS outperform the individual
classifiers. In situations, where the MCS is
constructed from heterogeneous classifiers, single
classifiers are deemed preferable.
3 PROPOSED METHODOLOGY
The proposed system (Figure 1) constitutes of the
following modules: data gathering, label encoding,
data type conversion, null value imputation, model
building and performance analysis.
Figure 1: Proposed methodology for the instance-based
classifier.
3.1 Data Gathering
In the study, 10 datasets are selected and these
datasets are divided into 3 main groups based on the
types of features present in the dataset. First group
contains datasets having only categorical features,
second group datasets contains only numerical
features and the last group contains datasets having
both categorical and numerical features. Dataset
selected in the first group are mushroom, car
evaluation and nursery dataset. Mushroom dataset has
been collected from the UCI repository and it
includes data on mushrooms which have been
labelled as either edible or poisonous. It consists of 23
features, including class label and have 8124
instances. Car Evaluation dataset has also been
extracted from UCI repository and is a multiclass
dataset. It includes data on car acceptability which has
been labelled as unacceptable, acceptable, good, and
very good condition. It has 6 features including class
label and is comprised of 1728 instances. Nursery
dataset has also been collected from UCI repository
and is a multiclass data set. It includes data to help
classify nursery school admission applications, and
the target has been recommended, priority, and
special priority admission. In total, it consists of 8
features, including class label and has 12960
instances. Dataset selected in the second group are red
wine, glass identification, and Pima Indians diabetes
dataset. Red wine dataset has been collected from
UCI repository and it has data which helps to classify
as good or not good. It comprises of 12 features
including target feature and the continuous value of
target variable of the dataset has been converted to
discrete by assuming score of wine greater than 7 as
good and rest as not good. Glass identification is a
multiclass dataset collected from UCI repository and
has data to help classify the type of glass based on
features. It has 9 features and constitutes of 214
instances. Pima Indians Diabetes dataset is a binary
class dataset that has been gathered using Kaggle. It
includes data which helps to identify whether a
person gets diabetes within five years of the first
AI4IoT 2023 - First International Conference on Artificial Intelligence for Internet of things (AI4IOT): Accelerating Innovation in Industry
and Consumer Electronics
436
physical examination. It constitutes of 8 features and
768 instances. Dataset in the third group includes
gender classification, car insurance, from Kaggle and
is a binary class dataset. It includes data that helps to
classify the gender of an individual. The dataset
consists of 8 features out of which 5 are categorical
features, 2 are numerical and 1 is the target class and
it has a total of 5001 instances. It has 19 features out
of which 7 are numerical features and 12 are
categorical features, including target variable, and
constitutes of 10000 instances. Lung cancer
prediction dataset has been collected from Kaggle and
it includes data which helps to classify the levels of
lung cancer. It is a multiclass dataset with target
variable being classified as high, medium and other.
It has 26 features, out of which 3 are numerical
features and 23 being categorical features, including
target variable. It comprises of 1000 instances. HR
Analytics dataset has been gathered from Kaggle
which includes data to help predict whether the
employee is looking for a job or not and the target has
been classified into binary values. It has 14 features
out of which 3 variables are numerical and other 11
are categorical, including the target variable. It
comprises of 19158 instances. Label Encoding is a
technique for transforming categorical data into
numerical data. In this paper, we have used label
encoding to convert the categorical variables into
numerical values. We have used it because we have
applied machine learning models that require
numerical inputs like Logistic Regression, KNN and
SVC. Label encoding also helps to conserve the
memory requirement for the processing of dataset, as
generally most of the categorical variables are defined
in a string or object format, which takes more space
and processing time as compared to numerical ones.
By reducing the memory required to contain
categorical variables, label encoding enables machine
learning models to analyze larger datasets and more
complex models with limited computational
resources.
3.2 Data Type Conversion
Data type conversion, also known as type casting, is
the process of transforming the data type of a variable
or value to a different type. Type conversion of the
feature is necessary before applying the model on the
dataset, as the proposed model performs different
operations on categorical and numerical features, due
to which it is necessary to convert categorical features
into dtype “category”, before applying the model. All
the other Machine Learning models do not require an
explicit dtype conversion to perform classification on
the data.
3.3 Null Value Imputation
Null value imputation is used to fill in missing or null
values in a dataset. It includes replacing the missing
values with estimated or imputed values and can be
done using a variety of methods. Numerical features
are usually imputed by finding mean of the non-
missing values of the column and categorical features
are imputed by finding the mode of the non-missing
values of the column. Imputed null values are
required, because they can increase model accuracy
by filling in missing values and preventing the loss of
crucial data. Algorithms like decision trees are more
robust to the presence of null values as compared to
Logistic Regression. Presence of null values can
affect the KNN algorithm, as it calculates the distance
between data points, which can be interfered with by
null values. If null values are not handled correctly,
they may result in bias or errors. Generally Median
imputation is used, when there are significant outliers
in the datasets. Since there are no significant outliers
present in the dataset, mean imputation has been
employed for the datasets containing missing values.
3.4 Model Building
Algorithm 1: Proposed instance-based classifier.
SimpleLearning (trainingDataSet, test):
class
For each attributeValue in test
For each class
If present in lookup
Get classCount
Else
For each instance in trainingDataSet
If attributeValue missing
Ignore
Else
Get classCount Increment
classCount
Endfor
Update lookup
Endfor
Endfor
class = max (attributeValue) // leading in more attributes
If tie
class = max (totalCount) // leads in total occurrences
across all attributes
If tie
class = any (class)
Model is built as per the Algorithm1 given below.
Model 1, 2 and 3 are based on the Algorithm1
Development and Comparative Analysis of an Instance-Based Machine Learning Classifier
437
mentioned below. Model 1 is implemented only for
categorical attributes, while model 2 deals with mixed
datasets. Model 3 also deals with mixed datasets but
incorporates dictionary and lookup features to
improve the execution time, which is not there in
models 1 and 2.
4 TIME COMPLEXITY ANALYSIS
It is obvious from the given algorithm that the time
complexity is O (d*c*n), where d is number of
attributes, c is the number of classes and n is the
number of instances (training). Table 1 lists time
complexity of various models used in our study
where, k: number of neighbours, T: number of trees.
Table 1: Model time Complexity.
Algorithm
Time Complexity
K-Nearest Neighbour
O(k*n*d)
Logistic Regression
O(n*d)
Decision Tree
O(n*log(n)*d)
Gradient Boosting
O(T*n*(log(d))
Proposed model
O (d*c*n)
Table 2: Performance on categorical datasets by various models is shown.
Dataset Name
Models
Accuracy (in %)
Precision
Recall
F1-Score
Mushroom dataset
(Binary
class)
Logistic Regression
95.63
0.97
0.94
0.955
KNN
99.8
0.99
1
0.99
Decision Trees
98.15
0.97
0.99
0.98
Gradient Boosting
100
1
1
1
Model 1
89.66
0.99
0.79
0.88
Model 2
89.66
0.99
0.79
0.88
Model 3
89.53
0.99
0.79
0.88
Car Evaluation
(Multi
class)
Logistic Regression
68.79
0.69
0.69
0.69
KNN
93.64
0.94
0.94
0.94
Decision Trees
89.6
0.9
0.9
0.9
Gradient Boosting
96.24
0.96
0.96
0.96
Model 1
73.41
0.73
0.73
0.73
Model 2
73.41
0.73
0.73
0.73
Model 3
73.41
0.73
0.73
0.73
Nursery dataset
(Multi
class)
Logistic Regression
77.16
0.77
0.77
0.77
KNN
93.6
0.94
0.94
0.94
Decision Trees
87.35
0.87
0.87
0.87
Gradient Boosting
98.69
0.98
0.98
0.98
Model 1
55.43
0.55
0.55
0.55
Model 2
55.43
0.55
0.55
0.55
Model 3
42.36
0.42
0.42
0.42
Table 3: Performance on numerical datasets by various models is shown.
Dataset Name
Models
Accuracy (in %)
Precision
Recall
F1-Score
Execution Time (in secs)
Red Wine dataset
(Binary
class)
Logistic Regression
89.38
0.67
0.26
0.37
16.176
KNN
89
0.61
0.28
0.39
0.019
Decision Trees
88.44
0.54
0.31
0.39
0.010
Gradient Boosting
90.94
0.68
0.49
0.57
0.338
Model 2
87.5
0.88
1
0.93
5.665
Model 3
87.5
0.88
1
0.93
6.774
Glass dataset
(Multi class)
Logistic Regression
55.81
0.56
0.56
0.56
0.350
KNN
69.77
0.7
0.7
0.7
0.010
Decision Trees
67.44
0.67
0.67
0.67
0.009
Gradient Boosting
72.09
0.72
0.72
0.72
0.733
Model 2
46.51
0.46
0.46
0.46
0.595
Model 3
46.51
0.46
0.46
0.46
0.625
Diabetes dataset
(Binary class)
Logistic Regression
74.68
0.77
0.46
0.58
0.035
KNN
67.53
0.61
0.38
0.47
0.013
Decision Trees
71.43
0.67
0.47
0.55
0.011
Gradient Boosting
72.73
0.7
0.48
0.57
0.221
Model 2
68.18
0.65
0.33
0.44
1.883
Model 3
68.18
0.65
0.33
0.44
2.015
AI4IoT 2023 - First International Conference on Artificial Intelligence for Internet of things (AI4IOT): Accelerating Innovation in Industry
and Consumer Electronics
438
Table 4: Performance on mixed datasets by various models is shown.
Dataset Name
Models
Accuracy (in %)
Precision
Recall
F1-Score
Execution Time (in secs)
Gender dataset
(Binary class)
Logistic Regression
96.6
0.95
0.97
0.96
0.056
KNN
97.1
0.97
0.97
0.97
0.047
Decision Trees
96.4
0.97
0.95
0.96
0.015
Gradient Boosting
97.2
0.97
0.97
0.97
0.303
Model 2
93.7
0.89
0.986
0.94
8.1
Model 3
93.7
0.89
0.986
0.94
7.563
Car Insurance
Dataset
(Binary class)
Logistic Regression
82.5
0.76
0.65
0.7
0.085
KNN
79.65
0.69
0.65
0.67
0.185
Decision Trees
83.75
0.74
0.75
0.746
0.025
Gradient Boosting
85.8
0.8
0.73
0.77
1.071
Model 2
68.2
0.68
1
0.81
42.584
Model 3
68.2
0.68
1
0.81
25.05
Lung Cancer
Prediction
(Multi
class)
Logistic Regression
100
1
1
1
1.308
KNN
100
1
1
1
0.085
Decision Trees
100
1
1
1
0.030
Gradient Boosting
100
1
1
1
1.027
Model 2
91.5
0.915
0.915
0.915
3.121
Model 3
91.5
0.915
0.915
0.915
0.677
HR Analytics
(Binary
class)
Logistic Regression
76.17
0.58
0.25
0.35
0.636
KNN
70.82
0.35
0.17
0.23
0.191
Decision Trees
78.24
0.6
0.43
0.5
0.034
Gradient Boosting
78.63
0.61
0.44
0.51
1.474
Model 2
74.47
0.74
1
0.85
76.89
Model 3
74.47
0.74
1
0.85
46.448
5 EXPERIMENTAL FINDINGS
In this study we have incorporated four machine
learning models which are KNN, Decision Tree,
Logistic Regression, and Gradient Boost. KNN is
used in the study as it is based on lazy learning
approach and is simple and doesn’t make any
assumptions about the distribution of the data.
However, it can be computationally expensive for
large datasets and is also sensitive to the choice of k.
Logistic regression was used as it is easy to
implement and understand relationship between
features and target variables. Decision Tree was
implemented in the study as it can handle non-linear
relationship in data and doesn’t make any assumption
about the distribution of data. However, it is prone to
overfitting and doesn’t work well on small datasets.
Gradient Boosting model was used in the study as it
can handle non-linear relationships and is less prone
to outliers. However, it needs tuning of hyper
parameters and can also overfit if the model is too
complex. As we can see from the tables 2 and 4 the
execution time for the model-3 has always been less
compared to the model 1 and 2. Model 3 is having a
reduced time largely because of the reason that it
creates a dictionary and it updates the test data of
majority vote of each attribute in the dictionary, due
to which after some test points the dictionary will
mostly have all the majority vote values for each
attribute and it is not needed to calculate the majority
vote of the attributes again, hence the testing time
gradually decreases . From the above tables existing
machine learning models like Logistic Regression,
KNN, Decision Tree and Gradient Boosting and
sometimes we can see the models outperforming the
KNN and logistic regression models, but we can also
observe the dip in the precision values because of the
class imbalance. When there is class imbalance in the
dataset then the algorithm votes for is high in number
because of which the minority class predictions are
outnumbered by the majority classes. Due to which
there is a decrease in correct classification of the
minority classes. All the algorithms are executed in
the Google Collab environment which has 12GB
RAM.
6 CONCLUSIONS
The results obtained from the implementation of the
proposed three model classifiers have been
comparable to those from the base line classifiers. In
some cases, the proposed model has performed better
while in other cases there has been some dip in
performance metrics due to the fact of imbalance and
high bias of output classes of datasets. From the
performance metrics, it has been observed that for
some datasets, where there is high bias towards one
Development and Comparative Analysis of an Instance-Based Machine Learning Classifier
439
class of output, the model under performs in terms of
performance metrics precision, recall and F1 score. In
the third model, there has been an improvement in
terms of computational time taken for training the
model. Another important observation is significantly
higher recall of the proposed classifier in certain
cases, which may need further investigation. Even
precision of proposed model is also noteworthy in
certain cases as compared to other models. One of the
future prospects of this paper could be to perform
various data sampling techniques to prevent the
model from over-fitting. Although the training time
of the model three has been considerably dropped,
there is still a scope for improvement in
computational time complexity of the model, by
leveraging various parallel architectures. In future,
the custom models can be tested on higher dimension
datasets as well.
REFERENCES
“UCI Machine Learning Repository.” n.d.
Archive.ics.uci.edu.https://archive.ics.uci.edu/dataset/
73/mushroom.
“UCI Machine Learning Repository.” n.d.
Archive.ics.uci.edu. Accessed March 17, 2023.
https://archive.ics.uci.edu/dataset/19/car+evaluatio
“UCI Machine Learning Repository.” n.d.
Archive.ics.uci.edu. Accessed March 17, 2023.
https://archive.ics.uci.edu/dataset/76/nursery.
“UCI Machine Learning Repository.” n.d.
Archive.ics.uci.edu.https://archive.ics.uci.edu/dataset/
186/wine+quality.
“UCI Machine Learning Repository.” n.d. Archive.ics.uci.
edu. Accessed June 17, 2023. https://archive.ics.uci.
edu/dataset/42/glass+identification.
“Pima Indians Diabetes Database.” n.d.
Www.kaggle.com.https://www.kaggle.com/datas
ets/uciml/pima-indians-diabetes-database
“Gender Classification Dataset.” n.d.
Www.kaggle.com.https://www.kaggle.com/datasets/el
akiricoder/gender-classification-dataset.
“Car Insurance Data.” n.d. Www.kaggle.com.
https://www.kaggle.com/datasets/sagnik1511/car-
insurance-data.
“Lung Cancer Prediction.” n.d. Www.kaggle.com.
https://www.kaggle.com/datasets/thedevastator/cance
r-patients-and-air-pollution-a-new-link.
“HR Analytics: Job Change of Data Scientists.” n.d.
Www.kaggle.com.https://www.kaggle.com/datasets/a
rashnic/hr-analytics-job-change-of-data-
scientists?select=aug_ train.csv.
Bi, Jingjun, and Chongsheng Zhang (2018). "An empirical
comparison on state-of-the-art multi-class imbalance
learning algorithms and a new diversified ensemble
learning scheme" Knowledge-Based Systems 158: 81-
93.
Caruana, Rich, Nikos Karampatziakis, and Ainur
Yessenalina (2008). "An empirical evaluation of
supervised learning in high dimensions" In Proceedings
of the 25th international conference on Machine
learning, pp. 96-103.
Zhang, Chongsheng, Changchang Liu, Xiangliang Zhang,
and George Almpanidis (2017). "An up-to-date
comparison of state-of-the-art classification
algorithms" Expert Systems with Applications 82: 128-
150.
A. Singh, N. Thakur and A. Sharma (2016) "A review of
supervised machine learning algorithms" 3rd
International Conference on Computing for Sustainable
Global Development (INDIACom), New Delhi, India,
pp. 1310-1315.
Caruana, Rich, and Alexandru Niculescu-Mizil (2006). "An
empirical comparison of supervised learning
algorithms" In Proceedings of the 23rd international
conference on Machine learning, pp. 161-168.
Brighton, Henry, and Chris Mellish (2002). "Advances in
instance selection for instance-based learning
algorithms" Data mining and knowledge discovery 6:
153-172.
Trivedi, Saksham, Balwinder Kaur Dhaliwal, and Gurpreet
Singh (2021). "A review paper on a comparative study
of supervised learning approaches" In 2021
International Conference on Computing Sciences
(ICCS), pp. 95-100. IEEE.
Al Makdah, Abed Al Rahman, Vaibhav Katewa, and Fabio
Pasqualetti (2019)."A fundamental performance
limitation for adversarial classification" IEEE Control
Systems Letters 4, no. 1: 169-174.
Gangwar, Vaishali (2012)."An overview of classification
algorithms for imbalanced datasets" International
Journal of Emerging Technology and Advanced
Engineering 2, no. 4: 42-47.
Sáez, José A., Mikel Galar, Julián Luengo, and Francisco
Herrera (2013)."Tackling the problem of classification
with noisy data using multiple classifier systems:
Analysis of the performance and robustness"
Information Sciences 247: 1-20.
Sarker, I. H. Machine Learning: Algorithms, Real-World
Applications and Research Directions. SN COMPUT.
SCI. 2, 160 (2021). https://doi.org/10.1007/s42979-
021-00592-x
Tammy Jiang, Jaimie L. Gradus, Anthony J. Rosellini,
Supervised Machine Learning: A Brief Primer,
Behavior Therapy, Volume 51, Issue 5, 2020, Pages
675 687, ISSN 0005-7894,
https://doi.org/10.1016/j.beth.2020.05.002.
(https://www.sciencedirect.com/science/article/pii/S00
05789420300678
AI4IoT 2023 - First International Conference on Artificial Intelligence for Internet of things (AI4IOT): Accelerating Innovation in Industry
and Consumer Electronics
440