Nidhika Yadav

Abstract: Decision Trees (DT) are a familiar and important tool in AI applications. One must understand, that the use of DT extends much beyond its current scope. Hence futuristic applications are proposed in article. That is to say, the AI toolkit can be used in a variety of applications. The current article explains Decision Tree form a business perspective and also lays importance to it’s inner details that a business consultant may need, in order to sell a software using this functionality, or to get a bid, of it, in some key proposed technologies that would be framed on it as base.The understanding for use, implementation and result computations on DT based methodologies are not very difficult, given one have time, patience and toolkits for the same.

Contents

Decision Trees and Futuristic Applications

Businesses today have become cut-short fast result based for demo of applications they wish to make. Well, anyone would want a cut-short demo of meaningful results that are so generated and provided. The reason is computing the outlook of solutions of the problems one may try to find. These solutions are required by business or research groups or even in academia. How nice and relaxing it may feel when a presentation of the implementation team provided insight to know-hows of the procedure to the board of the group, in charge of the problem solving. This way, the best, methodologies are presented, partial solutions are depicted, apart form future implementations, given, many people who want to get these implemented are themselves acquainted with the basics, at the minimum.

Decision Tree based algorithms, are one of the most explanatory and simple in details. The key idea of the simple approach of Decision Tree algorithm are as follows:

1. Simple to understand

The format of rules generated are simple for human understandings. Given somebody informs what the feature names and target names are.

2. Simple to understand as a tree (a simple graph)

The tree is the pictorial representation of the rules and is intuitive given basic understanding of how to traverse this graph with no cycles

3. Illustrative

Looking at the decision tree, once can see it is illustrative, given someone informs the stakeholder, what these arrows and numbers mean.

4. Predictions can be followed by non-technical person as well once the tree available

5. Workflow on synthetic data can be shown

This can be done, to show the steps involved. This is needed as business may want to calculate what may be charges and cost on change of datasets, any editing in data or so on.

6. Can to be taken to next level,

Once the initial bidding is over for a protocol, using real data, inferring and judgements, can be done.

7. Not complex to be transformed on new data which may be different but mostly compatible

This is required, given the data is also an expensive and pivotal part of a business. Fitting in, choosing the parameters is an essential part of Decision Trees for real time data. Very often use on data mining tasks.

The implementation can be considered as understanding Decision Tree on its own. Decision Trees require a parameter, perhaps a quantity, tomeasure whether a particular attribute can be taken as the next node of the decision tree. If so, what are the numerical or categorical points that separates the node into as pure classes as possible. Note, pure classes are those set of data points which are of same target class of this supervised algorithms. This is the basis of Decision Tree algorithm, explained in step below.

Steps of Decision Tree Algorithm

Step 1. Take the data, use pre-processing, depending of what data your algorithm is consuming as of now.Check some formats and so.

Step 2. Compute the Entropy, Information Gain or even Gini Index, you can experiment on these for the best results.

Step 3. Execute the algorithm, detailed python implementation shall be provided in separate article.Compute the node which is best, divide the data based on the parameter value.

Step 4. Get the decision tree, predict new entries, new data points. Go back to Step 3.

Step 5. Accuracies, results can be computed given the model is built.

This is how a Decision Tree works, sample decision tree computed on synthetic data is as follows.

Figure 1. This image is for illustration of how a Decision Tree looks like

This article talked about a high-level perspective of Decision Trees as a classification algorithm. There are immense applications of Decision Trees. However, the following are novel applications which one can work on in using Decision Trees algorithm. Moreover, one must know, some constraints in implementations. Future and many beneficial applications of Decision Trees, all these can be explained in detail in coming articles.

1. Normal Classification and Prediction tasks, these are mundane business tasks performed by everyone. These, I am not going in details of, it would be considered in implementation article.

2. Business Key Decisions. Can we even get a data when recession came in and can we get the parameters which led to it, can we then modulate and avoid recession altogether?Answer should be yes.

3. Climate Change adverse effect detection. The adverse effect per region can be computed with a DT.

4. Planet Next, to get tactics for next planet to visit, grow, DT can be used to perform computations.

5. Medical Global Guidelines: DT can be used to give complete overview of medical conditions in case of global or local necessities. More shall be explained with examples.

6. Trades of business, given businesses are volatile

7. Best Routes of Ships, using old data, to determine the best path, given glaciers are melting, sea level is rising

8. Robotics can use DT as often given substantial data to make decisions are given. Example a robot can itself decide to go out, or to take active part in an activity, given the readings of a sensor.

9. Probable best place to find rare metals, given equilibrium is needed to perform mining, if one place mining is closed another place it may be opened, for that a DT can be used to get the probable answers, before any more complex algorithm is utilized. Such kind of data needs people’s opinions, given the people who live there are best judge of it. Maybe we never need mining, perhaps.