Machine Learning Algorithms
Machine learning algorithms
• Data is the main resource for
machine learning models to identify
hidden patterns about business
processes and transactions.
• The performance of the machine
learning models is largely
dependent on the characteristics
and nature of data.
Types of Real-World Data
Structured:
• Organized and mostly represented in tabular format e.g. relational databases
Unstructured
• Unorganized having no predefined format such as pdf files, emails, wikis, web
pages etc.
Semi-structured
• It contains certain organization of data but not well structured. For example:
HTML and JSON documents can be considered as semi-structured.
Metadata
• Metadata gives the description about data. Author, file size, etc. of a document are Metadata.
Big data
• Sources of Big data can be
• The characteristics of “Big data” are: • Mobile devices
• Massive • IoT
• Social networks
• High dimensional
• Multimedia
• Heterogeneous
• Other applications
• Unstructured
• Unique features are (5Vs)
• Complex
• Volume
• Noisy • Velocity
• Incomplete • Variety
• Erroneous • Veracity
• Value
Types of Machine Learning Techniques
Types of Machine Learning Techniques
Supervised
• Supervised learning is the approach of finding relations between target or
dependent variables and certain set of inputs called independent variables.
• Classifications models performs predictions for certain class in target
variable while regression models find the best fitted line.
Unsupervised
• Unsupervised learning models work on unlabeled datasets to find the
related patterns from data to categorize data points to certain segments.
Types of Machine Learning Techniques
Semi-supervised
• Semi-supervised learning is the combination of both supervised and
unsupervised learning.
• It is used to label the unlabeled data depending on the result of supervised
learning.
Reinforcement
• The agent is trained on the basis of reward and penalty while interacting with
the environment.
• It leads to automation of sophisticated systems like robotics, automatic car
driving etc.
Supervised Machine Learning algorithms
• Classification algorithms
• Naive Bayes (NB)
• Logistic regression (LR)
• K-nearest neighbors (KNN)
• Support vector machine (SVM)
• Decision tree (DT)
• Random forest (RF)
• Adaptive Boosting (AdaBoost)
• Extreme gradient boosting (XGBoost)
• Regression Analysis
• Simple and multiple linear regression
• Polynomial regression
• LASSO and ridge regression
Unsupervised Machine Learning algorithms
• Cluster Analysis
• K-means clustering
• Mean-shift clustering
• Agglomerative hierarchical clustering
• Association Rule Learning
• Apriori
• FP-Growth
• ABC-RuleMiner
Dimensionality Reduction
Dimensionality reduction is a kind of unsupervised learning approach which reduces
models complexity by handling overfitting and thus allows lower computational costs.
• Feature selection
• Finding subset of unique features from set of attributes or variables.
• It eliminates irrelevant features to reduce the model complexity.
• Increases the speed of model training.
• Chi-squared test, ANOVA test, Pearson’s correlation coefficient are the examples of
feature selection technique.
Dimensionality Reduction
• Feature extraction
• Refers to the process of extracting new important features from existing
old features and thus eliminating the old features.
• New extracted set of features carries the majority of the information in the
dataset.
• Principal components analysis (PCA) is commonly used as a
dimensionality reduction technique.
General structure of a predictive model
Applications of Machine Learning
• Intelligent decision-making
• Threat intelligence
• Internet of things (IoT)
• Traffic prediction
• Healthcare
• Product recommendations
• NLP
• Computer vision
• Agriculture
• Context-aware applications
Challenges
• Collection of reliable data
• Handling noisy real-world data
• To clean and pre-process the diverse data
• Developing a model for target application is challenging as the
performance of the model varies on the characteristics of data.