The intent of this machine learning tutorial is to give a basic idea to the reader about machine learning. This is a very basic tutorial for beginners and intermediate professionals.
The entire machine learning tutorial is divided as mentioned below:
- History
- Overview
- Classification
- Conclusion
History of Machine Learning:
The term machine learning goes back to 1952 where a program could learn while running, which was created by Arthur Samuel. The first neural network was designed by Frank Rosenblatt, which was called perceptron and determined patterns and shapes.
Tom M. Mitchell gave the widely accepted definition for machine learning “A computer program is said to learn from experience E concerning some class of tasks T and performance measure P if its performance at tasks in T, as measured by P, improves with experience E.”
There was a phase where there was not much advancement in the area due to reduced funding and was given a term as “AI Winter”.
Machine Learning Overview:
Machine learning has become a new domain to look out for in this century. With its ever-growing impact on the various sectors, it is undoubtedly going to be crucial for shaping the next world. It is a subset of Artificial intelligence.
Machine learning is giving enough power to the machine to perform specific tasks, which we as humans are doing at the moment with minimal or no human intervention. So, it is letting machines learn with time and trial to come up with the desired output.
We, as humans, have also evolved similarly. If we pick up a particular task that we excel at this moment, it has resulted from various trials and errors, even basic actions that we perform today, that is, walking or speaking, is a consequence of numerous trials and errors.
Machine Learning takes previous data to predict future outcomes, or instead, it reads our behaviour and performs the way any human would perform at that moment.
Machine Learning Classification:
Machine learning could be divided into:
- Supervised learning- Supervised learning deals with problem statements where we know the output. Basically, here we have certain defined input variables that produce a specific outcome. So, when we have a new set of input variables, we need to predict the outcome. It is further subdivided into:
- Classification- As the name suggests it is used to classify into different outcomes, maybe ‘0’ or ‘1’. This could be used to classify into mails as SPAM or not, in the medical field to classify if cancer is malignant or benign etc.
There could be several other applications where the outcomes could be more than two. A typical outcome example could be determining the weather if it will be sunny, rainy or cloudy etc.
Several algorithms that are used for classification are logistic regression, Naïve Bayes, K-Nearest neighbors, Decision Tree, Random forest, Support Vector Machine.
- Regression- This is when your output is real value and we use the input parameters to reach or predict this value. A typical example could be determining the price of a house given input parameters as no of bedrooms, size, age etc.
Several algorithms used for this are linear regression, logistic regression, ridge and lasso regression, Decision tree regression, Random forest, K nearest neighbor, Support Vector Machine
- Unsupervised learning – This is when the machine analyzes the data and finds the patterns and forms a cluster. It deals with mostly unlabeled data that is there is no predefined outcome. Market segmentation is one such example where we form clusters based on input parameters so that we could predict when a new person comes to which cluster he/she belongs to. It could be further subdivided into:
- Clustering- It is grouping objects with the most similarities. Examples being grouping images of animals in different clusters, that is if we have a total of 50 images having 20 images of dog and 30 images of cat then by using clustering we will group dogs in one cluster and cat in another.
- Association- This is basically finding relationships between variables in a dataset. Example being if someone purchases bread then he/she is likely to purchase jam/butter.
Algorithms that are used for unsupervised learning are K-means clustering, hierarchical clustering, anomaly detection, principal component analysis, singular value decomposition.
- Reinforcement learning- This is based on rewards and penalties. Let’s say we wish to reach from point A to point B. At every step we will put a reward and penalty that the machine takes. Now the intention of the machine is to maximize the reward. A very common example of this is self-driving cars wherein the machine learns by itself how to react in a particular situation.
Conclusion:
This brings us to the end of the machine learning tutorial. Machine learning was introduced many years ago, but it has made significant advancements in the last ten years. Its involvement in various spheres allows the human race to make much more substantial progress than it has created ever before. From its involvement in healthcare to indicate the start of the disease, it sure is interesting to see how and when this would take up an entire world.
