Modelling and Evaluation

Rule-Based and Learning-Based Models

Reason — The two primary types of modelling techniques in AI are rule-based and learning-based models.

Rule-Based Models

Reason — Rule-based modelling is a way to make decisions using a list of specific rules. These rules are predefined and followed step by step, that is, they are set beforehand by humans.

Ability to handle unexpected scenarios

Reason — A key disadvantage of rule-based modelling is that the system can only handle situations covered by the predefined rules, and therefore unexpected scenarios can be challenging.

Root Node

Reason — The root node represents the initial point of the decision tree where the first decision occurs.

Learning from mistakes

Reason — Reinforcement learning is also referred to as learning from mistakes, as the machine learns through the process of trial and error.

Precision and recall

Reason — F1 Score is the harmonic mean of precision and recall, providing a balance between the two metrics.

1. Learning-based learning models improve over time by learning from data, without following strict predefined rules.
2. Supervised learning requires a large amount of labelled data, which can be time-consuming and expensive to collect.
3. Reinforcement learning is often compared to the human learning process of trial and error.
4. Evaluation metrics like accuracy and F1 score are used to assess the performance of an AI model.
5. In a decision tree, the terminal points where final outcomes are determined are called leaf nodes.
6. Edge deployment approach processes data directly on devices like IoT sensors or smartphones.

A is true but R is false.

Reason — Learning-based models learn from data and improve over time, eliminating the need for manual rule updates. However, being limited to predefined rules and not able to modify behaviour without manual updates is a characteristic of rule-based models, not learning-based models.

1. Making the model accessible for general usage and timely updates.
   Reason — Deployment of a model refers to making the model accessible for general usage and its timely updating to achieve the desired result.

2. It ensures the model is used by the intended users and updated timely.
   Reason — Deployment is important because the AI model becomes available for real-world use and can be updated to achieve the desired result.

3. General usage by students, library staff and faculty members.
   Reason — The library assistant application can only be useful if it is deployed so that it can be accessed by the students, library staff and faculty members.

4. Model deployment.
   Reason — After learning the various evaluation techniques for a model, the next phase is to deploy the model, making it accessible for general usage.

Decision nodes are locations within a decision tree where decisions are required, resulting in further branches based on the outcome of the decision.

In the prediction phase, once the model is trained, new input data is given to the model and it predicts the output based on what it learned during training.

Leaf nodes are the terminal points of a decision tree where final decisions or outcomes are determined.

In edge deployment, the AI model is deployed directly on devices like smartphones, sensors or IoT devices, where data is processed near its source to provide quick results with low latency and offline capability.

F1 Score is the harmonic mean of precision and recall, providing a balance between the two metrics used to evaluate the performance of an AI model.

Regression is a type of statistical and machine learning technique used to model the relationship between a dependent variable (target) and one or more independent variables (predictors). It is a form of supervised learning that uses labelled datasets to predict continuous outputs, such as house prices, sales growth or weather conditions.

True Positive (TP) symbolises that the model correctly predicted the output. For example, if an image contains a cat and the model correctly detects the cat, it is a true positive.

True Negative (TN) is when the model rightly identifies a negative case. For example, if a model correctly recognises that there is no cat in an image without a cat, it is a true negative.

Evaluation is one of the most crucial phases in the AI project cycle. It assesses how well an AI model performs its intended task and helps ensure that the model is reliable, accurate and effective in real-world applications. The primary objective of evaluation is to identify the model's strengths and weaknesses and decide whether it is ready for deployment or needs further improvement.

ROC-AUC is used to evaluate the performance of a classification model. The ROC (Receiver Operating Characteristic) curve is a graphical representation that shows the relationship between the True Positive Rate (TPR) on the y-axis and the False Positive Rate (FPR) on the x-axis at different threshold values. AUC (Area Under Curve) is a single scalar value that summarises the overall performance of the ROC curve and shows how well the model can distinguish between positive and negative classes.

Differences between cloud deployment and on-premises deployment of AI models are:

The advantages of Edge Deployment are as follows:

• Low Latency: It provides real-time results as data does not need to be sent to the cloud for processing.
• Offline Capability: It can function without an internet connection.

Advantages of Rule-Based Modelling:

1. It is easy to understand, as the system’s decisions are clear and follow straightforward rules.
2. It is simple to create, since setting up a rule-based system only requires defining rules.
3. It is consistent, as the system behaves predictably by always following the same rules under the same conditions.

Disadvantages of Rule-Based Modelling:

1. It is not flexible, as the system can only handle situations covered by predefined rules.
2. It is difficult to manage, because maintaining and updating the system becomes harder as the number of rules increases.
3. There is no learning, as the system does not learn from new experiences and uses the same rules unless manually updated.

The different phases of a learning-based AI model are:

1. Training Phase
2. Learning Phase
3. Prediction Phase

Data is important in the training phase because the learning-based model analyses the given data to learn patterns and features, which helps it build a model that can make accurate predictions. The quality and quantity of data directly affect how well the model learns and performs.

The key characteristics of supervised learning are as follows:

1. It learns from labelled data, where the correct output is already known.
2. The model is trained using input data and corresponding output provided by a supervisor.
3. It is used to analyse labelled datasets and discover hidden patterns to arrive at an output.
4. Once trained, the model can predict correct answers for new or similar input data.

Advantages of Supervised Learning:

1. Supervised learning often provides very accurate results because the model is trained on labelled data, meaning it knows the correct answers during training.
2. Supervised learning is easier to predict how the model will perform because it learns from data that has clear and correct labels.

Disadvantages of Supervised Learning:

1. Supervised learning requires a large amount of labelled data, which can be time-consuming and expensive to collect.
2. Sometimes, the model might become too focused on the training data and not perform well on new, unseen data. This is called overfitting.

The various techniques used in the supervised learning process are:

1. Regression Techniques
2. Classification Techniques
3. Neural Network Techniques

One real-world application of supervised learning is email spam detection. In this application, supervised learning models classify emails as spam or not spam based on labelled examples of past emails. The model learns from emails that are already marked as spam or not spam and then uses this learning to predict whether new incoming emails are spam or legitimate.

Differences between Supervised Learning and Unsupervised Learning:

Differences between Positive Linear Regression and Negative Linear Regression relationship are:

Linear regression is a supervised machine learning algorithm that establishes a linear relationship between a dependent variable and one or more independent features by fitting a linear equation to the observed data.

The benefits of using the techniques of linear regression are as follows:

1. It is easy to understand, implement and interpret.
2. It is computationally efficient and suitable for small to medium-sized datasets.
3. The coefficients provide clear insight into the relationship between input variables and the output.
4. It effectively models relationships where the dependent and independent variables have a linear association.
5. It is useful for both academic and practical applications like trend analysis, forecasting and predictive modelling.

Any three real-world applications of linear regression are:

1. It can estimate the value of a property based on features like location, size and age of the house.
2. It is an efficient tool to analyse historical sales data and predict future sales trends.
3. It can predict healthcare expenses based on patient demographics and medical history.

The challenges associated with unsupervised learning are as follows:

1. Complexity: Interpreting the results of unsupervised learning can be difficult, as there are no clear labels or outcomes to guide the analysis.
2. Uncertain accuracy: Without labelled data, it is challenging to evaluate how accurate or effective the model is, making it harder to measure success.
3. Requires domain knowledge: Understanding and validating the patterns discovered by unsupervised learning often require expert knowledge in the specific field.
4. Risk of overfitting: Since the model tries to find patterns in all the data, it might sometimes identify patterns that are not meaningful, leading to overfitting.
5. Not always suitable: Unsupervised learning may not be the best approach for tasks where specific outcomes or predictions are needed.

Reinforcement learning differs from supervised and unsupervised learning in the way the model learns. In supervised learning, the model learns from labelled data under guidance, while in unsupervised learning, the model discovers patterns in unlabelled data without supervision. In contrast, reinforcement learning learns through trial and error, where an agent interacts with the environment, learns from its mistakes, and improves its actions by maximising rewards over time.