Supervised, Non-supervised Semi-supervised ML? Which method is right for you?
It is widely accepted that there are two main categories of machine learning - Supervised and Non-supervised. However, we believe a third very important category should be defined - the semi-supervised method of machine learning should be set as a separate category. With this sub-type, you are achieving a much more reliable result within a given context, with some pretty significant differences to both the supervised and non-supervised learning types.
Let’s take a look at each sub-type in more detail.
Supervised Machine Leaning: This method is employed for the label data. It is effectively used for Genetics and Deep Learning types of algorithms. For instance, Convolutional Neural Network Algorithm can be trained with a specific purpose and they are quite effective in doing that, if you set the target correctly and train it based on the well-designed data set. It is important to understand and define the result you are trying to achieve in order to design the training data set. Otherwise the algorithm will be achieving the goal as the training set defines it. Overfitting and under-fitting are the definitions used to describe malfunctions for the Machine Deep Learning Algorithms. Where the overfitting is more preferable issue to have as there is a number of techniques that can be employed to deal with it. Where the under-fitting will be an issue as you will not be able to get the required performance for the unlabelled data after the training.
Non-Supervised Machine Leaning: If you have a data that is aggregated into a pool and not-labelled, then you can employ the algorithms that are be able to identify patterns to distinguish clusters by dissimilarities between the data set without any labelling provided. Decision trees will be able to classify the data into many classified clusters that are formed on some similar properties.
Semi-Supervised Machine Leaning: There are classification and clustering algorithms that can be used with the various degree of certainty. For instance, you can have clustering for the unlabelled data that was pooled within aggregated context and this can form a semi-supervised technique for k-means, decision trees, binary categorisation etc. These algorithms are also used for unsupervised clustering.