Machine Learning Techniques

Unit-I: Introduction Machine Learning (Short

Question)
Q1. Define machine learning.
Ans. Machine learning is an artificial intelligence application that allows
computers to automatically learn and improve from experience without being
explicitly designed.

Q2. What are the different types of machine learning algorithm ?

Ans. Different types of machine learning algorithm are:
 1. Supervised machine learning algorithm.
 2. Unsupervised machine learning algorithm.
 3. Semi-supervised machine learning algorithm.
 4. Reinforcement machine learning algorithm.

Q3. What are the applications of machine learning ?

Ans. Applications of machine learning are:
 1. Image recognition
 2. Speech recognition
 3. Medical diagnosis
 4. Statistical arbitrage
 4. Statistical arbitrage

Q4. What are the advantages of machine learning ?

Ans. Advantages of machine learning:
 1. Easily identifies trends and patterns.
 2. No human intervention is needed.
 3. Continuous improvement.
 4. Handling multi-dimensional and multi-variety data.
Q5. What are the disadvantages of machine learning ?
Ans. Disadvantages of machine learning :
 1. Data acquisition
 2. Time and resources
 3. Interpretation of results
 4. High error-susceptibility

Q6. What is the role of machine learning in human life ?

Ans. Role of machine learning in human life:
 1. Learning
 2. Reasoning
 3. Problem solving
 4. Language understanding

Q7. What are the components of machine learning system ?

Ans. Components of machine learning system are :
 1. Sensing
 2. Segmentation
 3. Feature extraction
 4. Classification
 5. Post processing

Q8. What are the classes of problem in machine learning ?

Ans. Classes of problem in machine learning are :
 1. Classification
 2. Regression
 3. Clustering
 4. Rule extraction

Q9. What are the issues related with machine learning ?

Ans. Issues related with machine learning are :
 1. Data quality
  2. Transparency
 3. Tracability
 4. Reproduction of results

Q10. Define supervised learning.

Ans. Supervised learning, also known as associative learning, involves training the
network by feeding it input and matching output patterns.

Q11. Define unsupervised learning ?

Ans. Unsupervised learning, often known as self-organization, is the process by
which an output unit is trained to respond to clusters of patterns in the input.

Q12. Define well defined learning problem.

Ans. A computer programme is said to learn from experience E with respect to
some class of tasks T and performance measure P if its performance at tasks in T,
as measured by P, improves with experience E.

Q13. What are the features of learning problems ?

Ans. Features of learning problems are :
 1. The class of tasks (T).
 2. The measure of performance to be improved (P).
 3. The source of experience (E).
Unit-II: Decision Tree Learning (Short Question)
Q1. Define decision tree learning.
Ans. Decision tree learning is a predictive modelling approach used in statistics,
data mining, and machine learning. It use a decision tree to progress from
observations about an item to conclusions about the item’s desired values.

Q2. What is decision tree ?

Ans. A decision tree is a decision-making tool that employs a tree-like model of
decisions and their potential outcomes, such as chance event outcomes, resource
costs, and utility.

Q3. What are the types of decision tree ?

Ans. There are two types of decision tree:
 1. Classification tree
 2. Regression tree

Q4. Define classification tree and regression tree.

Ans. Classification: A classification tree is an algorithm with a fixed target
variable. This algorithm is used to determine which class a target variable belongs
to.
Regression tree: A regression tree is an algorithm that predicts the value of a
variable whose value is not fixed.

Q5. Name the decision-tree algorithm.

Ans. Decision-tree algorithms are :
 1. ID3
 2. C4.5
 3. CART
Q6. What are the issues related with the decision tree ?
Ans. Issues related with decision tree are:
 1. Missing data
 2. Multi-valued attribute
 3. Continuous and integer valued input attributes
 4. Continuous-valued output attributes

Q7. What are the attribute selection measures used in decision tree?
Ans. Attribute selection measures used in decision tree are:
 1. Entropy
 2. Information gain
 3. Gain ratio
 4. Gini index

Q8. Define artificial neural network.

Ans. Artificial Neural Networks (ANN) or neural networks are computational
algorithms designed to mimic the behaviour of biological systems made up of
neurons.

Q9. What are the advantages of ANN ?

Ans. Advantages of ANN are:
 1. Easy to use.
 2. Alter to unknown condition.
 3. It can model difficult function.
 4. It can be imposed in any application.

Q10. What are the disadvantages of ANN?

Ans. Disadvantages of ANN are:
 1. Hardware dependence.
 2. Unexplained functioning of the network.
 3. Assurance of proper network structure.
 4. The difficulty of showing the problem to the network.
  5. The duration of the network is unknown.

Q11. Name different types of neuron connection.

Ans. Different types of neuron connection are:
 1. Single-layer feed forward network.
 2. Multilayer feed forward network.
 3. Single node with its own feedback.
 4. Single-layer recurrent network.
 5. Multilayer recurrent network.

Q12. Define gradient descent.

Ans. Gradient descent is an optimisation technique used in Machine Learning and
Deep Learning that can be used to most, if not all, learning algorithms.

Q13. What are the different types of gradient descent ?

Ans. Different types of gradient descent are :
 1. Batch gradient descent.
 2. Stochastic gradient descent.
 3. Mini-batch gradient descent.

Q14. Define ADALINE.

Ans. ADALINE is a single linear unit Adaptive Linear Neuron network. The delta
rule is used to train the Adaline network. It receives input from numerous units as
well as the bias unit. Adeline models are made up of trainable weights. The inputs
are either positive or negative, while the weights have a sign (positive or
negative).

Q15. Define backpropagation algorithm.

Ans. The backpropagation method is a supervised learning training algorithm for
feedforward neural networks. Backpropagation computes the gradient of the loss
function with respect to the network weights quickly for a single input-output
case.

Q16. What is perceptron ?

Ans. The perceptron is the most basic type of neural network utilised for
categorization of linearly separable patterns. It is made up of a single neuron with
variable synaptic weights and bias.

Q17. What is multilayer perceptron ?

Ans. Multilayer perceptrons are perceptrons that are stacked in layers. This model
is composed of three layers: an input layer, an output layer, and a hidden layer.

Q18. What are the parameters that affect the backpropagation neural
network?
Ans. Parameters that affect backpropagation neural networks are :
 1. Momentum factor
 2. Learning coefficient
 3. Sigmoidal gain
 4. Threshold value

Q19. What are the selection parameters used in BPN (Backpropagation

Neural Network)?
Ans. Selection parameters used in BPN are:
 1. Number of hidden nodes
 2. Momentum coefficient α
 3. Sigmoidal gain 𝛌
 4. Local minima
 5. Learning coefficient η

Q20. What are the ANNs used for speech and character recognition ?
Ans. ANNs used for speech recognition are :
 1. Multilayer network.
 2. Multilayer network with recurrent connections.
ANNs used for character recognition are :
 1. Multilayer neural network such as backpropagation neural network.

Q21. What are the advantages of mini-Batch gradient descent ?

Ans. Advantages of mini-Batch gradient descent are :
 1. Easily fits in the memory.
 2. It is computationally efficient.
 3. It performs vectorization.
Unit-III: Evaluating Hypotheses (Short Question)
Q1. Define hypothesis.
Ans. In supervised learning, a hypothesis is a function that describes the target. A
hypothesis is a preliminary association between two or more variables that guides
study.

Q2. What are characteristics of hypothesis ?

Ans. Characteristics of hypothesis are :
 1. Empirically testable
 2. Simple and clear
 3. Specific and relevant
 4. Predictable
 5. Manageable

Q3. What is the importance of hypothesis ?

Ans. Importance of hypothesis are :
 1. It given a direction to the research.
 2. It specifies the focus of the researcher.
 3. It helps in devising research techniques.
 4. It prevents from blind research.
 5. It ensures accuracy and precision.
 6. It saves resources, time, money and energy.

Q4. What are different types of hypotheses?

Ans. Different types of hypotheses are :
 1. Simple hypothesis
 2. Complex hypothesis
 3. Working hypothesis
 4. Alternative hypothesis
 5. Null hypothesis
 6. Statistical hypothesis
 7. Logical hypothesis
Q5. What are the difficulties faced in estimating the accuracy of hypotheses ?
Ans. Difficulties faced in estimating the accuracy of hypotheses are :
 1. Bias in the estimate
 2. Variance in the estimate

Q6. What are different methods of sampling ?

Ans. Different methods of sampling are :
 1. Simple random sampling
 2. Systematic sampling
 3. Stratified sampling

Q7. Define Bayesian decision theory.

Ans. Bayesian decision theory is a fundamental statistical method to pattern
recognition. This method is built on quantifying the tradeoffs between various
classification judgements by utilising the likelihood and costs associated with such
decisions.

Q8. Define Bayesian belief network.

Ans. Joint conditional probability distributions are specified using Bayesian belief
networks. Belief Networks, Bayesian Networks, and Probabilistic Networks are
other names for them.

Q9. Define EM algorithm.

Ans. The Expectation-Maximization (EM) algorithm is an iterative method for
determining maximum-likelihood estimates for model parameters when the data is
incomplete, has missing data points, or contains hidden variables.

Q10. What are the usage of EM algorithm ?

Ans. Usage of EM algorithm are :
  1. It can be used to fill the missing data in a sample.
 2. It can be used as the basis of unsupervised learning of clusters.
 3. It can be used for the purpose of estimating the parameters of Hidden
Markov Model (HMM).
 4. It can be used for discovering the values of latent variables.

Q11. What are the advantages of EM algorithm ?

Ans. Advantages of EM algorithm are :
 1. It is always guaranteed that likelihood will increase with each iteration.
 2. The E-step and M-step are easy implementation.
 3. Solutions to the M-steps exist in the closed form.

Q12. What are the disadvantages of EM algorithm ?

Ans. Disadvantages of EM algorithm are:
 1. It has slow convergence.
 2. It makes convergence to the local optima only.
 3. It requires both the probabilities, forward and backward (numerical
optimization requires only forward probability).
Unit-IV: Computational Learning Theory (Short
Question)
Q1. Define computational learning theory.
Ans. Computational Learning Theory (CLT) is an Al research topic that studies the
design of machine learning algorithms to discover what types of problems may be
learned.

Q2. What are the uses of computational learning theory ?

Ans. Uses of computational learning theory (CLT) are :
 i. It provides a theoretical analysis of learning.
 ii. It shows when a learning algorithm can be expected to succeed.
 iii. It shows when learning may be impossible.

Q3. Define sample complexity.

Ans. Sample complexity is the number of training samples that must be supplied to
the method in order for the function given by the algorithm to be within an
arbitrarily tiny error of the best feasible function, with probability arbitrarily close
to 1.

Q4. What the variants of sample complexity ?

Ans. There are two variants of sample complexity :
 1. Weak variant: The weak variant fixes a particular input-output
distribution.
 2. Strong variant: The strong variant takes the worst-case sample
complexity over all input-output distributions.

Q5. What are the parameters on which complexity of a learning problem

depends ?
Ans. Complexity of a learning problem depends on :
 1. Size or expressiveness of the hypothesis space.
  2. Accuracy to which target concept must be approximated.
 3. Probability with which the learner must produce a successful hypothesis.
 4. Manner in which training examples are presented, for example, randomly
or by query to an oracle.

Q6. Define principal component analysis.

Ans. Principal Component Analysis (PCA) is a statistical process that employs an
orthogonal transformation to convert a set of correlated variables to a set of
uncorrelated variables.

Q7. Define mistake bound model of learning.

Ans. An algorithm A learns a class C with mistake bound M iff Mistake (A, C) ≤
M.

Q8. What is instance-based learning ?

Ans. Instance-Based Learning (IBL) is a classification algorithm that is an
extension of nearest neighbour or KNN classification algorithms that does not keep
a collection of abstractions of models produced from instances.

Q9. What are the advantages of KNN algorithm ?

Ans. Advantages of KNN algorithm are :
 1. No training period.
 2. Since the KNN algorithm does not require training before making
predictions, new data can be supplied without affecting the system’s
accuracy.
 3. KNN is easy to implement.

Q10. What are the disadvantages of KNN algorithm ?

Ans. Disadvantages of KNN algorithm are:
 1. It is does not work well with large dataset.
  2. It is does not work well with high dimensions.
 3. It need feature scaling.
 4. It is sensitive to noisy data, missing values and outliers.

Q11. Define locally weighted regression.

Ans. Locally Weighted Regression (LWR) is a memory-based approach that
performs regression around a point of interest using local training data.

Q12. Define radial basis function.

Ans. A Radial Basis Function (RBF) is a real-value function that assigns a real
value to each input from its domain, and the value produced by the RBF is always
an absolute value, i.e., it is a measure of distance and cannot be negative.

Q13. Define case-based learning.

Ans. Case-based learning algorithms take a sequence of training instances as input
and output a concept description, which may be used to predict goal feature values
for following cases.

Q14. What are the disadvantages of CBL (Case-Based Learning) ?

Ans. Disadvantage of case-based learning algorithm:
 1. They are computationally expensive due to the fact that they save and
compute similarities to all training cases.
 2. They are sensitive to noise and irrelevant characteristics.
 3. They are affected by the algorithm’s similarity function selection.
 4. They cannot process symbolic valued feature values in a straightforward
manner.

Q15. What are the functions of CBL ?

Ans. Functions of case-based learning algorithm are :
 1. Pre-processor
  2. Similarity
 3. Prediction
 4. Memory updating

Q16. What are the processing stage of CBL ?

Ans. Case-based learning algorithm processing stages are :
 1. Case retrieval
 2. Case adaptation
 3. Solution evaluation
 4. Case-base updating

Q17. What are the benefits of CBL as lazy problem solving method ?
Ans. The benefits of CBL as a lazy Problem solving method are:
 1. Ease of knowledge elicitation.
 2. Absence of problem-solving bias.
 3. Incremental learning.
 4. Suitability for complex and not-fully formalised solution spaces.
 5. Suitability for sequential problem solving.
 6. Ease of explanation.
 7. Ease of maintenance.

Q18. What are the applications of CBL ?

Ans. Applications of CL:
 1. Interpretation
 2. Classification
 3. Design
 4. Planning
 5. Advising

Q19. What are the advantages of instance-based learning ?

Ans. Advantages of instance-based learning:
 1. Learning is trivial.
  2. Works efficiently.
 3. Noise resistant.
 4. Rich representation, arbitrary decision surfaces.
 5. Easy to understand.

Q20. What are the disadvantages of instance-based learning ?

Ans. Disadvantages of instance-based learning :
 1. Need lots of data.
 2. Computational cost is high.
 3. Restricted to x ∊ Rn.
 4. Implicit weights of attributes (need normalization).
 5. Need large space for storage i.e., require large memory.
 6. Expensive application time.
Unit-V: Genetic Algorithm (Short Question)
Q1. Define genetic algorithm.
Ans. Genetic algorithms are search and optimisation algorithms that are based on
the mechanics of natural genetics and natural selection. To design search and
optimisation procedures, these algorithms replicate the principles of natural
genetics and natural selection.

Q2. Give the benefits of genetic algorithm.

Ans. Benefits of genetic algorithm are :
 1. They are Robust.
 2. They provide optimization over large space state.
 3. They do not break on slight change in input or presence of noise.

Q3. What are the applications of genetic algorithm ?

Ans. Following are the applications of genetic algorithms :
 1. Recurrent neural network
 2. Mutation testing
 3. Code breaking
 4. Filtering and signal processing
 5. Learning fuzzy rule base

Q4. What are the disadvantages of genetic algorithm ?

Ans. Disadvantages of genetic algorithm :
 1. Identification of the fitness function is difficult as it depends on the
problem.
 2. The selection of suitable genetic operators is difficult.

Q5. Define genetic programming.

Ans. Genetic Programming (GP) is a subset of machine learning that is a form of
Evolutionary Algorithm (EA). EAs are used to find solutions to problems that
humans are unable to address.
Q6. What are the advantages of genetic programming ?
Ans. Advantages of genetic programming are :
 1. In GP, the algorithm can generate an enormous number of potential
programmes.
 2. Although GP employs machine code, which speeds up results, any high-
level language that is utilised must be compiled, which can cause errors and
slow down our programme.
 3. There is a strong likelihood that even little variations have a devastating
influence on the fitness of the created solution.

Q7. What are the disadvantages of genetic programming ?

Ans. Disadvantages of genetic programming are :
 1. It does not impose any set solution length, hence the maximum length can
be extended up to hardware restrictions.
 2. It is not necessary for an individual to have complete knowledge of the
problem and its solutions in genetic programming.

Q8. What are different types of genetic programming ?

Ans. Different types of genetic programming are :
 1. Tree-based genetic programming
 2. Stack-based genetic programming
 3. Linear genetic programming
 4. Grammatical evolution
 5. Cartesian Genetic Programming (CGP)
 6. Genetic Improvement Programming (GIP)

Q9. What are the functions of learning in evolution ?

Ans. Function of learning in evolution :
 1. It enables individuals to adapt to changes in the environment that occur
over the course of a person’s life or through several generations.
 2. It enables evolution to employ information gathered from the
environment, directing evolutionary search.
 3. It can help and guide evolution.
Q10. What are the disadvantages of learning in evolution ?
Ans. Disadvantages of learning in evolution are :
 1. A delay in the ability to acquire fitness.
 2. Increased unreliability.

Q11. Define learnable evolution model.

Ans. Learnable Evolution Model (LEM) is a non-Darwinian evolutionary
computation model that uses machine learning to drive the development of new
individuals (candidate issue solutions).

Q12. What are different phases of genetic algorithm ?

Ans. Different phases of genetic algorithm are :
 1. Initial population
 2. FA (Factor Analysis) fitness function
 3. Selection
 4. Crossover
 5. Mutation
 6. Termination

Q13. Define sequential covering algorithm.

Ans. A general approach that continually learns a single rule to generate a decision
list (or set) that covers the entire dataset rule by rule is known as sequential
coverage.

Q14. Define Beam search.

Ans. Beam search is a heuristic search strategy that expands the most promising
node in a constrained collection to explore a graph.

Q15. What are the properties of heuristic search ?

Ans. Properties of heuristic search are:
 1. Admissibility condition
 2. Completeness condition
 3. Dominance properties
 4. Optimality property

Q16. What are different types of reinforcement learning ?

Ans. Different types of reinforcement learning are:
 1. Positive reinforcement learning
 2. Negative reinforcement learning

Q17. What are the elements of reinforcement learning ?

Ans. Elements of reinforcement learning are :
 1. Policy (𝝅)
 2. Reward function (r)
 3. Value function (V)
 4. Transition model (M)

Q18. Define Q-learning.

Ans. Q-learning is model-free reinforcement learning, and it is often easier to
implement. Reinforcement learning is the challenge faced by an agent that must
learn behaviour through trial-and-error interactions with a dynamic environment.

Q19. Define positive and negative reinforcement learning.

Ans. Positive reinforcement learning :
 a. Positive reinforcement learning happens when an event occurs as a result
of a certain behaviour, such as increasing the strength and frequency of the
activity.
 b. In other words, it has a positive effect on the behaviour.
Negative reinforcement learning: Negative reinforcement is described as the
strengthening of a habit as a result of a negative circumstance being ended or
avoided.