From automating mundane tasks to offering intelligent insights, companies in every sector are talking about Machine Learning as it is driving and creating tangible business value for a diverse range of industries, but only if used in the right way.
Machine Learning Algorithms and its diverse capabilities and platforms are helping lots of industries to develop new business models, improve the quality of their products and services and optimise their operations. For example:
- Amazon uses machine learning to offer personalised services to its customer based on purchase history.
- Siri uses machine learning and deep learning to understand human language.
- Facebook started recognising similar faces.
- Google introduced smart reply function to Gmail for quick reply.
Even though Machine Learning is very topical, still many don't have a true understanding about what Machine Learning and AI is all about. This post is for those who are curious about machine learning and AI and are wondering about its magic.
Machine Learning is a subfield of AI that gives system the ability to learn and improve from experience. AI is a branch of computer science which emphasise on making smart machines to think and work like humans. In traditional programming, data was used for writing programs in order to get the desired output. In Machine Learning, there are few generic algorithms that show interesting facts about the data. For these algorithms, we don't need to write any specific code, data plays a key role here and based on the type of data, the algorithms make its own logic. For example, Classification algorithm, which classifies the data into various groups. Few examples are:
- to classify text based on different emotions (positive, negative, neutral).
- to classify email as ham or spam.
- to classify articles about sports, movies, politics etc.
Supervised Machine Learning
In supervised machine learning technique, the dataset should always be labelled. There is a relationship between input and output. The problem you solve using this technique is to predict the output/ labels for data points without a label e.g. classifying email as ham or spam. The learning model also compares its output with the correct output and improve accordingly.
Unsupervised Machine Learning
In unsupervised machine learning technique, the dataset is not labelled. This algorithm can't figure out the right output, but it divides the data into various clusters and structures having different properties.
Semi-Supervised Machine Learning
Semi-supervised falls in-between supervised and unsupervised machine learning as it uses both labelled and unlabelled data for training.
Reinforcement Machine Learning
It is a kind of machine learning algorithm in which the machine and software agents get a delayed reward for its previous action in the next step. To maximize agent's performance, this method helps in automatically identifying the ideal behaviour in specific context. Mostly used in games e.g. Mario.
Data Collection step is the first and most important step in machine learning. In machine learning data is very important. An excellent quality and quantity data with a decent algorithm gives far better results than a powerful algorithm with less data. The better the quality of the data, the better the learning process of machine learning will be.
2. Data Preparation
This phase includes all the activities to process the data to make the final dataset for the modelling. This is also known as Data pre-processing. It includes data analysis, data cleaning, variable selection, fixing issues like missing data and handling outliers etc.
3. Training the Model
This step includes finding the best algorithm and data representation in the form of model. In this step the data is divided into two parts- training set and test set. Training set is to train the model and test set is to evaluate the model.
4. Evaluating the Model
Next is the evaluatation of the model before final deployment and involves examining of all the steps to make sure that the algorithm and approach is 100% correct. After this the next step is to check the accuracy of the outcome.
5. Improving the Performance
Finally, a key component is improving the model further by feature engineering or by introducing more variables.
Machine Learning helps to analyse huge amounts of structured and unstructured data. Machine learning is very useful in getting faster and accurate results. Today, in this competitive world, businesses are not only satisfied with the solution to problems like: did we generate the required revenue for this year, did we achieve our sales target or did we target the correct customers etc., a huge emphasis is to know the likelihood in the future and the probable outcomes and Machine learning is definitely a potential solution for this.