1 - intro Flashcards
what is management science?
applying scientific models to improve decision making
optimisation
data-based, empirical, objective, quantitative.
aim to analyse, predict and control
what is analytics?
using quantitative methods to derive meaning from data to make informed business decisions
what is modelling?
translating reality into mathematical and logical formulations such that it can analyse the status quo, test alternative possible scenarios and generate insights
what is the tradeoff when making a good model?
tradeoff between simplicity and accuracy. complex models need reliable, available data which is not common. even if you have it, will you be able to address the more complex model? how much improvement is expected from the complex model, and is it outweighed by the costs? diminishing returns on adding detail.
occam’s razor
what is descriptive analysis?
stats, simulation, machine learning. used to give simple picture of system, identify trends and bottlenecks etc. model system in analytical way to identify relationships between components. usually followed by predictive or prescriptive steps.
what is descriptive modelling?
simulation - see how the system works. can tell average queue times, utlisation rates of machines etc.
statistical analysis - descriptive stats. correlations, regression.
machine learning - e.g. finding demand functions.
what is predictive analysis?
forecasting and time series analysis, regression, machine learning, game theory. predicts what’s going to happen in future. forecasting demand, which customers have high risk of default, will there be a housing market crash etc.
what statistical methods can be used for predictive analysis?
time series analysis - examining and modelling patterns, trends and behs in sequential data collected over time. moving average, exponential smoothing, ARIMA, VAR. variants use exogenous variables.
regression analysis and choice modelling - e.g. linear regression, logistic regression, choice models, probit models.
how can game theory and behavioural economics help with predictive analysis?
e.g. predicting russia/ukraine war, gov negotiations, competing companies in markets
beh economics - frameworks to understand and predict behaviour of individuals, e.g prospect theory.
how can machine learning help with predictive analysis?
machine learning - most important nowadays. advanced techniques to find relationships between input and output variables. proper when complex relationship between input/output. examples of predicting stock prices, too many inputs for a regression. general structure input > function > output. ANN vs DNN.
decision trees and random forests - can be used for training models, random forests used when there are many decision trees with different weights. useful for things like shortlisting job applicants.
good for prediction power, flexibility and pattern recognition. however data intensive, complex, computation resource demand and interpretability.
what are some challenges with predictive analytics?
data - availability, sufficiency, cost of acquisition, reliability, bias.
complexity and interpretability
computational efforts
reliance on big data
potential bias in results of analysis
what is prescriptive analysis?
optimisation (linear/ mixed integer/nonlinear, Stochastic, Robust Dynamic), Heuristic approaches, Game theory, Decision analysis approaches, Simulation. all about deciding what to do to achieve certain objective.
what is optimisation?
optimise objective given set of constraints. can be static or dynamic. dynamic means things like optimising pricing for seats on certain flight up to 3 months before flight dates. Deterministic models vs. models to deal with ambiguity, risk and uncertainty (e.g., robust optimisation, stochastic programming), etc. optimisation techniques stop you from having to consider every possible solution, saving time.
what is a heuristic approach to optimisation? what are some examples?
These approaches aim to solve the optimization problems with lower computational efforts. They do not guarantee finding the optimal solutions, but they aim for a near-optimal solution with small optimality gap
Example of heuristic algorithm: Nearest Neighborhood for the Travelling Salesman Problem
what is a meta-heuristic algorithm?
Meta-heuristic algorithms are usually flexible, and can be applied on a wide variety of problems, e.g., resource allocation, routing, finding optimal location, batch sizing, etc