Core Methodologies of Machine Learning.

Machine learning space encompasses 2 significant tasks which aids in data analysis and output prediction. They are called Supervised Machine Learning and Unsupervised Machine Learning.

Supervised Machine Learning works on labeled data, which is fed as the input, to generate a training model that is capable of making future predictions.

Unsupervised Machine Learning on the other hand, works on unlabeled data to identify the patterns/relationships/ hidden features in the given dataset.

• Labeled data - When each input has a corresponding output, it is termed as labeled data.
• Unlabeled data - When there is no predefined output for each input, it is termed as unclassified or unlabeled data.

Listed below are the major differences between Supervised Machine Learning and Unsupervised Machine Learning.-

Categories of Supervised Learning :

1. Classification - It is used to specify data into specific class. It predicts discrete labels or categories for the data points (Output is always categorical). It can be either binary classification or of multi-class classification.

2. Regression - It predicts continuous numerical values (Output is always a numerical value).

Categories of Unsupervised Learning :

1. Dimensionality Reduction - In this method, data is represented using lesser columns/features while preserving data integrity.

2. Clustering - Groups together unlabeled data according to "distance", i.e based on their similarities or differences.

3. Autoencoders - It encodes input distribution into a common pattern (representations) for all samples and then decode the representations back into input space. Here we use neural network for representation.

Use Cases -

Supervised Learning -

1. Spam email detection

2. Time series forecasting (stock price prediction, weather forecasting)

3. Object Recognition, Image Tagging. (OCR- optical character recognition)

Unsupervised Learning -

1. Customer segmentation

2. in the fields of Bioinformatics and Genetic engineering (DNA pattern clustering)

3. Recommendation system (news feed generation, relevant ads generation)

Pros and Cons :

Supervised Learning -

Pros:

1. We can use this model to predict future outcome based on some prior experiences.

2. This model gives us the accurate idea on the object classes.

3. Helps us address different real time problems such as spam detection, fraud identification, etc.

Cons:

1. Building the training models takes long computation time.

2. It cannot generate the right result if the data to be tested is different from the training model.

3. Not best suited for dealing with complex tasks.

Unsupervised Learning -

Pros:

1. It is comparatively easier to get unlabeled data.

2. Yields underlying, hidden pattern which is previous unknown.

Cons:

1. It is complex as compared to Supervised Learning.

2. Since the input dataset is not labeled/unclassified and we have no prior idea of the result to be generated, therefore the output generated by this algorithm may not be very accurate.

Note : It is important to pre-process the data in order to get rid of any anomalies (noise/outliers, null values, etc.)