The dynamics of Machine Learning
By Kevin Doshi, SE Comps
AI and Machine Learning have plowed their way into our digital lives in such ingenious yet sometimes-inconspicuous ways that we use their applications on a regular basis without realizing that they are making the wheels turn. Its applications include risk management in insurance and banking, predicting the onset of diseases, filtering spam messages in your e-mail, making recommendations related to products that you’ ve viewed or songs that you’ ve listened to, identifying new malware using older malware, in self-driving vehicles, showing more relevant search results and so on. Much of what we thought was difficult to do using computers is now practically possible thanks to the power of machine learning. Here is a summary on how it works:-
Machine learning is defined as methods or algorithms that are used to teach a computer to identify patterns in data and thus predict outcomes, without being explicitly programmed to do so. Let’ s take an example where we want to teach a computer to differentiate between tea and coffee. The first step in machine learning is one where a lot of human intervention is required- data gathering and preprocessing.
Assume that we have a dataset of tea and coffee, having 2 characteristics or‘ features’ – caffeine content, cost, the‘ target attribute’ i. e. whether the product is tea or coffee. Generally, the caffeine content per 100g of tea is about 10mg whereas that of coffee is about 40mg. Tea costs anywhere between 35 and 150, whereas coffee costs between 125 and 400. This becomes our initial assumption. Now, our dataset may be in excel, csv and may contain some null or incorrect values. Our first task is to remove those entries, make sure our data is ready for training the computer and set aside some 20-30 % data for evaluation. It is important to choose a proper training dataset, which reflects general behavior with some deviations.
Once we have our training data ready, we have to choose a ML model to employ. There are various models like binary classification, linear regression, naïve Bayes, clustering, nearest neighbor, decision trees and neural networks to choose from. Our choice depends upon the type of dataset and the required outcome. Since our problem definition is quite basic, we can use a simple linear model that will get the job done. The next step is to train our model. Let’ s say our model is an X-Y graph with a straight line representing our initial assumptions.
If our dataset contained an entry having 14mg caffeine and costing 70 ₹ , it would probably guess tea. Similarly, if an entry had 38mg caffeine and cost 200 ₹ it would probably