Deep Learning-Based Method for Detecting Cassini-Huygens Spacecraft Trajectory Modifications
Deep Learning-Based Method for Detecting Cassini-Huygens Spacecraft Trajectory Modifications
Fájlok
Dátum
Szerzők
Ashraf ALDabbas
Zoltán Gál
Aldabbas Ashraf Khaled Abd Elkareem (1983-) (informatikus)
Gál Zoltán (1966-) (informatika- és villamosmérnök)
Folyóirat címe
Folyóirat ISSN
Kötet címe (évfolyam száma)
Kiadó
Absztrakt
Supervised learning is a task of machine learning that maps an input to an
output based on available data. The data set contains the information from
which we get to know the process of classifying at least some of the data
correctly. Thus, the classified data is called Training set. In this learning
method, the supervision comes from the instances having labels in the training set. Classification problems are mostly referred to come from the branch
of supervised learning. Classification is a function of machine learning that
finds out the correctly predicted class labels of instances for all the unlabelled
instances. In this research, our main focus will be discussing about the most
commonly used data classification methods especially Naïve Bayes (NB) and
Boosting basically Adaptive Boosting (AdaBoost) classifiers. In preparation
for the enhancement of the performances with regards to the accuracy rate
of those classifiers we would like to introduce two newly hybrid approaches
for classification when the data set is big, noisy and high dimensional. In
real life, the data sets usually contain noise or outliers, contradictory instances or missing values. Mostly the data got affected by them during the
time of data collection or generation. To resolve that issue, we proposed two
new hybrid classifiers. Our first hybrid approach is the ADA+NB classifier,
where we used Adaptive Boosting (AdaBoost) classifier to find comparatively
more important attribute subsets before the assumption of class conditional independence using Naïve Bayes (NB) classifier [13]. Besides, NB classifier
assumes class conditional independence, therefore, it’s possible to multiply
the probabilities when the events are independent. Because of that, NB classifier can be very effective in removing examples from the training set before
the decision tree (DT) generation at the time of building AdaBoost model.
We named this process our second proposed hybrid NB+ADA classifier [14].
This paper investigates the comparison between two classical machine learning approaches with two new hybrid classifiers in terms of accuracy rate, error
rate, precision, f-score, sensitivity, specificity analysis on 4 real benchmark
data sets who are high dimensional and noisy chosen from UCI (University of
California, Irvine) machine learning repository. For instance, Adult data is
one of the renowned noisy data set available in UCI. Therefore, we tested AdaBoost classifier which gave us accuracy of 87.65% and NB classifier provided
79.99%. On the other hand, our first proposed (ADA+NB) classifier showed
86.39% and our second proposed (NB+ADA) indicated 94.14% of the accuracy rate with the same data. Similarly, we used other data-sets and derived
performance comparison between the classifiers to prove that our proposed
classifier’s performances are higher than the text-book classifier’s.
Leírás
Kulcsszavak
Forrás
Proceedings of the 1st Conference on Information Technology and Data Science / Fazekas István. -Debrecen : Creative Commons License, 2022. -p. 19-31