Michel chamat dimitrios bersi kodra

Clustering 
Visualization and dimensionality 
reduction 
Association 
rule learning 
o
k-Means 
o
Hierarchical 
Cluster Analysis 
(HCA) 
o
Expectation 
Maximization 
o
Principal Component Analysis 
(PCA) 
o
Kernel PCA 
o
Locally-Linear Embedding (LLE) 
o
t-distributed Stochastic Neighbor 
Embedding (t-SNE) 
o
Apriori 
o
Eclat 
Fig. 10. Comparison between supervised and unsupervised learning [27]. 
 
 
Supervised Learning
Unsupervised Learning
Classification
Regression
Clustering

---

23 
3.3. 
Restraints and Drawbacks 
As useful and more efficient ML is becoming in our lives, as dangerous it 
is if the drawbacks of this method are not taken into consideration. 
Following, the most significant restraints are presented and briefly 
explained. [21][23]
1. 
Massive Data Acquisition: 
ML algorithms require enormous amounts of data for training to 
perform their complex tasks that the developers designed them for. 
Also, a good quality restriction must be designated to get the best 
results. 
2. 
Resources: 
Need of extended resources in time for the algorithm training to fulfill 
the high standards of accuracy and relevancy. Computational power 
for the operation of these large amount of data must be considered 
likewise.
3. 
Result Comprehension:
The understanding of the ML outcomes can be tricky. Accordingly, 
the selection of the right algorithm should be done carefully to 
optimally interpret these results. 
4. 
Error Sensitive:
ML is hypersensitive to errors, for this reason, the training sets have 
to secure that the samples are unbiased and exclusive. However, it is 
difficult to achieve a good result and even more complicated to 
recognize the issue and correct it if the dataset includes a big amount 
of erroneous data.