A brief tour into the Machine Learning World
Part-1: — INTRODUCTION
“Machine Learning”, The name says everything about this terminology. It means a type of algorithm family which is intended to teach machine to perform what user want.
In now a days the data is continuously increasing day by day and so is the rise in resources to handle those data. So, to handle those one need a sophisticated algorithm which can handle those huge chunks of data without manual intervention and process it in minimal time. The analyst needed some algorithms which drives an insights from those gazillions of dataset and derive the useful facts from the same and then those facts will be used by business persons to take an effective decisions.
So, here to solve all these problems of handling such mammoth data and to derive the useful effective insights from the same, “Machine Learning” comes into picture. So, what machine learning is all about?
Machine Learning: — The Machine Learning is a part of AI (Artificial intelligence) that helps in developing the algorithmic model to handle the large dataset. The model derive by ML are developed by allowing the computer to learn from the data.
The model finds pattern in data and tries to predict the output based on the data insights. To rate the model derive by ML as efficient or effective the pattern derived from such model must be meaningful so that they lead to some advantage.
Studying ML algorithms will help us in developing our models and also help us in handling of our large dataset effectively in minimal time. The ML also help us in performing the high-level analysis on our dataset like prediction, regression, clustering, and much more. (I’ll try to cover all the aspect of ML including the advance Deep Learning in this series.)
So, now the question arises is how ML model work?
For working of any ML model, first we pre-process our data and clean it with outliers and then we need to train the model with the dataset. We feed some chunks of our dataset (70% data of dataset is normal practice) in form of training data set into the ML model. Then the ML model start iterating over the training data for several iteration. During the iteration it uses the concept of weight(w) and bias(b) on each iteration and modify the error incur during each iteration with these two values as [ new_value = old_value*weight + bias ]. By using this the data loss has been minimized during each iteration.
In next step, when training has been done then the model predict some output for each row of data and the output then tested with the remaining data other than training one (30% data of dataset in normal practice). So, after prediction done by ML model, test data has been fed to the testing code along with trained ML object and program decided what will be the expected predicted output class for each row. Now the predicted output class of each testing data row has been compared with the original class belong to that row and then the accuracy of algorithm has been derived.
If the accuracy comes too low then the ML model has been modified and re-trained with training data to improve the accuracy or still if the model’s accuracy doesn’t improve then we have to understand that the model we derived is not suitable for the data we have. So, we have to use another ML model in such case.
And if the model accuracy comes good then we will save the object of that model for future and again when it require to run the same model on same dataset then we don’t need to train the model again rather we will use the object of that model to predict the output.
In next part we will continue the further discussion over different types of ML model and data processing technique.
Stay tuned…. Thanks :)
