Analytics: Types of Analytics

Source: Introduction to Data Literacy on DataCamp

We are officially transitioning from Data Collection to Data Analysis with this section. We are going to discuss the types of Analytics

They are four types of Analytics:

1. Descriptive Analytics: What is Happening?
2. Diagnostic Analytics: Why is it Happening? (Root Cause)
3. Predictive Analytics: What will happen? (Probability of Possible Outcome)
4. Prescriptive Analytics: What should we do? (Best Course of Action for desired outcome)

As we go further down, the more advanced the question (analytics) becomes. This does not mean, the more advanced is the best choice.

If we can solve our problem with Descriptive Analytics — why use Prescriptive Analytics? It will just take more time and energy

You choose the analytics based on the question you want to solve

Descriptive Analytics

This is based on summarising and visualising the data. The leading question is: “What is happening?”

Here, we are just trying to get to know the data based on the variables. What are variables: These are characteristics of interest we have measured or observed For example: Top Speed in a car comparison dataset.

This is the analytics where we investigate the relationship or patterns in the data. For Example: When there is a specific grouping or the variable is closely related to another. It can be used as preparation exercise/tool for more advanced techniques like building a model.

At its most basic form: Descriptive Analytics consist of Descriptive Statistics like: Average/Maximum/Minimum values of variables. We can use this to answer questions like: How fast did we go? How slow did we go? What was our average speed?

Visualisation:

Visualising also helps us to see the characteristics of the data — to see patter, groupings and other relationships in the data.

It can also help with Outlier Detection. This helps us to see values that are differnt from the majority

The more advanced and comprehensive form of Descriptive Analytics is: Exploratory Data Analysis (or EDA)

Exploratory Data Analysis (EDA) Focuses on exploring the data:

• Assessing its main characteristics
• Finding relationship patterns or groups
• Suggesting hypothesis for future analysis: This is contrasted with Hypothesis Testing which is more formal and specific.

EDA looks at data in a broader, open way that allows us to discover something that is not know yet.

It combines different techniques to get a comprehensive picture of the data with a story emphasis on visualisation

It includes Basic Descriptive and more Advanced techniques such a cluster analysis to find groupings.

Descriptive Analytics is used as groundwork for further analysis or it can just be valuable on its own for identifying insights like discovering Data Problems

Diagnostic Analytics

Just like a doctor diagnosing a patient, Diagnostic Analytics focuses on finding the root cause of an event. It is similar to descriptive analysis but focuses more on the “Why is this happening?” question.

So why is this important: Diagnostic Analytics

• It help us find the potential cause of events or reason for behaviour
• Investigate casual relationships
• Most Importantly: You can suggest solutions based on identified causes

Important Note: While it is expected of Diagnostic Analytics to provide evidence of causal relationship — it is much tougher to prove causality on its own

Why: Statistical relationship between two variables can be caused by a variable we don’t know yet or even coincidence. This is why domain knowledge is so crucial in the aspect of Diagnostic Analytics.

Remember earlier when we said, Data-Driven Decision Making is not just the responsibility of the data team and it is more enriched when we have people from other teams as part of the process. This is one of the many reason why.

They can help us determine whether the results are plausible

On top of that, we can set up additional experiments and analysis to strengthen the case of casualties

Common Techniques

• Drill Down Analysis: We analyze the general summary by analysing the underlying data. For example, if you are examining global warming and you noticed that the earth temperature has decreased (General Data) — you can apply drill-down analysis to identify which location increased the most + the season why the temperature was the highest on earth to determine probable causes for why the earth’s temperature has increased. In better, you can investigate reasons why the earth temperature has increased the most.
• Correlation and Regressing Analysis: They tell us which variables are closely related
• Hypothesis Testing: In this case, the probable cause for an event is theorized before the technique is applied
• Root Cause Analysis (RCA): This is a formal set of steps that is used to look beyond the superficial causes that have a direct effect. Some of these methods use similar methods as descriptive analytics and rely on similar descriptive statistics and visualisation techniques

The key difference between Descriptive Analytics is that focuses on a broader view whilst diagnostic analytics is more linear in its focus on the underlying causes

Let’s expand on Root Cause Analysis:

These formal set of steps look beyond superficial cause that have a direct effect. These steps investigate the superficial causes (contributing factor) by investigating the factors further to identify their causes

This root causes analysis keeps drilling down until the root causes are identified. How far we can go on this is typically determined by domain knowledge

These are the 5 steps:

1. Define the event under consideration
2. Collect all relevant data related to this event
3. Determine the contributing factors based on the analysis of the data
4. Drill down on the root causes
5. Recommend solutions based on the identified root causes

This analytics also relies on domain knowledge and the results cna be used to recommend solutions for the most common categories of compliants

Predictive Analytics

Here there is a great divide:

• Descriptive and Diagnostic Analytics focuses on the Past and at most the Present
• Predictive and Prescriptive Analytics look at the future

Predictive Analytics are focused on identifying outcomes and their probability of happening. They help us investigate on what will most likely happen based on what we know from the data.

It can assist in anticipating the most likely outcome of an event — like predicting when a machine will fail based on maintenance data so companies can organise repairs in advance

It is also often sued to forecast how a specific trend will continue — E.g. predict stock price or evolution of pandemic

It is used to predict something that is not yet kjnow in the present (linked to future events)

• Example: Based on Symptoms — preduc which disease a patient most likely had
• Example: Based on the characteristics of a transaction, predict whether it is fraudulent or not
• Example: Based on the characteristics of a land, determine whether it is rich in oil or mineral deposits

Nonetheless, — predictions are associated with a degree of uncertainity and the uncertainty increases as we look further away from the data.

What are some common techniques:

1. Machine Learning Models: This is a technique when computer learn from existing data to identify existing patterns in the data to make predictions on the new data. This is what GMAIL uses to know when an email you just received might be a spam. A distinction can be amde between classification based and Regression based Models

Classification Based Models: This predicts categories or membership to a group like — e.g. Predicting whether a customer will cancel their subscription in the near future

Regression-Based Models: This predicts values — For example, like predicting Housing prices based on neighbourhood characteristics

Time-series Forecasting: Predicting future values over time

Predictive Text Analytics: This is a sub-group of predictive analytics that is focused on text, typically used to predict which category a text belong — Example: Whether an email is spam or not.

They are some common steps all predictive models use. This is the breakdown below

1. Define the Outcome
2. Collect and prepare the data: Typically this is split into two sets:
3. Training set for building the model
4. Test Set — a smaller set used for evaluating the model created on the Training Set.
5. Build the Model: These data are then used to build and train the model until it provides accurate prediction on the training data (Training Set).
6. Interpret and Evaluate the Model: When the model is ready, its predictors are evaluated on the test data and then assessed for accuracy and other pre-determined metrics (Multiple Metrics are often used to evaluate a predictive model thoroughly) And example of a pre-determined metrics could be % of correct predictions
7. Implement the model and fine-tune over time: Finally the model can further fine-tuned if necessary

Prescriptive Analytics

This builds on the results of Predictive Analytics. This analytics aims on taking it further and taking action on the predicted outcome. With prescriptive analytics. we are trying to answer the question: “What is the prescribed or recommended course of action?”

By doing this, we are selecting the best action given the outcome we want to achieve and we take into account the related benefits and drawbacks

The primary purpose of prescriptive analytics is to help us decide what best to do?

Prescriptive Analytics has the following benefits: 1. Make informed, data-driven decision 2. Optimise processes 3. Mitigation of risk associated with particular outcomes

Common Techniques are:

1. Rules Based System: involves generating a set of rules or decision logic to get best outcome. This rules are generated by domain knowledge, machine learning or combination of both

2. Re-enforcement Learning: An algorithm learns to achieve a particular objective or optimise an outcome by receiving positive or negative feedback when running through a set of actions.

3. Scenario and Simulation Analysis: Running through a set of pre-determined scenarios or simulating multiple outcomes to help select the decision that leads to the best outcome

4. Recommendation Engines: You see this when shoping online, looking at reels at IG, watching videos on YT (Right Side list of videos) or watching movies on Netflix. You are recommended what to buy next. Behind these recommendation are a recommendation engine. It looks at data on what products you viewed or purchased in the past and uses that as data to predict what other product you would be interested in purchasing. Providing specific recommendation on what you should purchase next based on predicted interests.