A solution given by a predictive model can be more reliable if it gets optimized for being a proper solution to the problem. Different approaches of machine learning are used to build predictive models whereas different approaches of operations research are used to find optimal solutions. The combination of both of these approaches gives such solutions which are not only accurate but also optimal. In this article, we are going to discuss the combination of machine learning and operation research and how it helps in solving specific problems where accurate and optimal solutions are needed. We will also discuss a few notable use cases of this combination. The major points to be covered in this article are listed below.
Table of Contents
Sign up for your weekly dose of what's up in emerging technology.
- What is Operations Research?
- Characteristics of Operations Research
- Uses of Operations Research
- Machine Learning in Operations research
- Way to Hybridization of ML and OR
- Comparing Operations Research and Machine Learning
- Example of Combination of OR and ML
- Solving problems of ML using OR
- Recommendation system
- Computer Vision
- Sentiment Analysis
- Use Cases of Combination of ML and OR
What is Operations Research?
Operation research is used as an analytical approach or method which can help in solving problems and making decisions. This decision and problem-solving approach can help in management and benefits of an organization. The basic approach for solving problems using operation research can start with breaking down the problem into basic components and ends with solving those broken parts in defined steps using mathematical analysis.
The overall procedure of operation research can be completed into the following steps:-
- Identification of the problem
- Constructing a model which can understand and resemble the real world and variables in the surrounding of the problem
- Use the model to make solutions to the problem
- Optimize all solutions provided by the model
- Implement the optimal solution to the problem
Concepts of operation research became very useful for the world during World War II because of the military planner. After the world war, these concepts have become useful in the domain of society, management, and business problems.
Characteristics of Operations Research
There are the following characteristics of a basic operations research procedure:-
- Optimization – The main goal of operation research is to find an optimal solution to the given circumstances. By optimization of a solution, we can find the best solution where it involves comparison and testing of the solution by operation research.
- Simulation – In simulation built models are used to try and test the solutions before applying them in the procedure.
- Probability and statistics – Mathematical algorithms and data help in making insights and checking risks which allow us to make reliable predictions and test possible solutions.
Uses of Operations Research
There are a variety of problem and decision-making domains where operations research can be helpful. Some of them are listed below as:
- Scheduling of tasks and management of their time.
- Planning of urbanization and agricultural steps.
- Supply chain management
- Enterprise resource planning
- Inventory and godown management.
- Risk management
- Network marketing.
By the above, we can say that the operation research approach is far better than ordinary software and data analytic tools. An experienced person in operation research can benefit an organization to achieve more complete datasets and using all possible outcomes can predict the best solution and estimate the risk.
The above image is a representation of the operation research procedure with its main components. We can say that operation research is a science of optimization using which we can obtain a huge number of improvements in any field. Some of the papers and research are examples of 20-40% of the improvement in the problem-solving domain.
Machine Learning in Operations Research
In the above section, we have an overview of the operation where we have seen how we can find an optimal and best solution to a problem and how we can make decisions using simple steps. When we talk about machine learning we can say the algorithms under machine learning work on the basis of learning from the past histories of the data and information under the data and the main motive of the algorithms is to predict an accurate value that can satisfy the user and perform the task accurately for which model is assigned.
We can say that OR and ML both work on finding the better solution to a problem where models in machine learning can also be used in making decisions. For experienced operation research things become difficult when the set of the solution becomes higher and manually performing the testing of the solutions becomes hectic and time taking. Also with this testing task the experienced need to estimate the risk before applying the solution to the problem of making any decision. Using machine learning we can reduce the time taken by the operation research and the manual iteration between the testing. Hybridization of ML and OR can be considered as the next advancement of operation research where models from machine learning can help in various tasks that come under operation research.
Way to Hybridization of ML and OR
We can perform the hybridization of ML and OR in the following four ways:-
- ML then OR – Here the ML can help in finding the points or the solution and then after using the OR we can optimize the points of solutions.
- ML in OR – Here we can say that ML is helping us to perform tasks that come under the OR. This can be considered as the operation research procedure.
- OR in ML – Here we can say that the operation research is helping in performing tasks of the machine learning procedures and this can be considered as the machine learning procedure.
- New Hybridization of ML and OR – In this, we can consider it as the total hybridization of ML and OR where we receive some new algorithms
Comparing Operations Research and Machine Learning
Let’s go through an example where we are in a city, let’s call it Mumbai and we want to travel around Mumbai in an optimal way so that we can cover the most number of locations in a short time and at less cost. So to do this using machine learning we are required to optimize all the possible ways and their times and cost so that the model related to the machine learning can predict an optimal way by considering all the facts in the account. When the same problem comes in the way of operation research it can be thinking of the cost or time or the distance and we can find more than one solution for the problem and after applying them all once we can find an optimal way.
By these procedures of both, we can say that the number of nodes and steps taken by the machine learning algorithms is less than the number of nodes and steps taken by the operation research. We can even say that many of the building blocks of the machine learning models are taken from the operation research procedure. Some of the examples are as follows:
- Gradient descent
- Bellman’s equations
- Euclidean distance.
Example of Combination of OR and ML
Let’s go through one more example of a road construction company which has got a tender from the government. The task of the company is to repair the road defects. This can be done by the combination of machine learning and operation research where the machine learning models can help in identifying the type of road defects like broken roads in a small area, medium area or large area After that, using the operation research, we can find the beneficial policies for replacement and repairing of the road. This can be a work procedure where the machine learning and operation research is used together for the development. Similarly, there are various domains where we are required to work on both of the technologies for approaching the solution to a problem in a better way.
Solving Problems of ML Using OR
The paradigm of machine learning can be considered as the combination of various domains like sentiment analysis, computer vision, and recommender systems where applying OR with them can help us in various aspects. Also, it can help in solving problems that occur with machine learning. Let’s talk about the problems of machine learning and how we can solve it using operation research.
- Recommendation System
As we know that recommendation systems are becoming more important for a lot of business domains because of their success in providing fruitful recommendations to the user of the business and using these recommendations the owner of the business can make a lot of benefits also they are made using the machine learning procedure where they are used for giving recommendations.
Let’s take an example of the restaurant where we have enabled services like online booking and machine learning algorithms are helping in estimating various aspects like eating time of the customer, habits of the customer and customer bookings and recommendation system are installed to provide recommendations to the users according to those attributes of the users. The problem with these instalments comes when the traffic of the customer is very high and the online booking system starts getting confused about the table allotment to the customer.
In such a situation operation research can help in increasing the traffic by managing them and system response time where the work of the operation research procedure can be optimizing the real-time booking, the number of people eating in the real-time, expected number of customers in a particular time. These optimizations can help in simulating the bookings with customer behaviour. This simulation can be done by combining the OR and ML together.
- Computer Vision
The computer vision algorithms of the machine learning paradigm work on the visual data and one of the main tasks of these algorithms are to classify or identify the images from a given set of images. Let’s say we have a computer vision algorithm to track the food demand on a similar restaurant. where a deep learning model is installed with cameras and working for estimating the food wastage and it is working by recognizing the food type and estimating the food demand.
Since we know that pixels of the images will be the main factor in which the classification is dependent and due to distance and size sometimes we face the failure of the deep learning models. An operation research procedure can be enabled with the machine learning or deep learning algorithm, where it can be used for tracking the different matching algorithms between the frames of the image and we can optimize the maximum number of food sold and amount of food wasted.
- Sentiment Analysis
In the field of sentiment analysis we know we have reached so far in the context of advancement and now many of the systems have become so reliable when we talk about the results that they are producing. One of the major problems with these systems or for making these systems we require a lot of data. And we know it is tough and costly to make such data available for the models. In this scenario, we can use operation research for optimizing data that can be accurate, effective, and cost-effective for the model.
Frequently it happens that the data we gather for modelling is biased by an emotion that can be estimated and tracked by the operation research. When we talk about the NLP system we know that the system cannot autonomously change its emotions and they are also allowed to control them less. Using the operation research we can make them controlled by just optimizing systems behaviour and results.
As we know that the machine learning models are based on the parameters which we need to fit in the models so that using the parameter and the data model be trained to perform the task which is assigned to the model and also we see that before feeding data into the model we require parameters that can help the model to work well with the data. Optimization of the parameters can be done by operation research because we have defined earlier that operation research is a science of optimization. The better fit parameters can be obtained by optimizing the sets of parameters using the operation research techniques.
Use Cases of Combination of ML and OR.
As of now, we have seen various ways and benefits of using the OR and ML together. In this section of the article, we will discuss some real-life use cases of this combination. Since both of them are very relatable to each other many of the big giant companies like google, amazon, etc. are using the combination to obtain a good result and provide customer satisfaction for example:
- Google Maps is an example of a combination of the OR and ML where MLis used to suggest the ways and predictions about the future traffic. Along with that, Google Maps also helps in finding the optimal way to reach the destination which is a very good example of the OR and ML combination.
- Amazon logistics uses the combination of big data, machine learning, and operation research for optimizing logistics. A detailed description of the combination is given in the following link.
- A startup named funarttech is focused on making a product in the field of NLP named as an open door for AGI is a system where OR is combined with the ML. using which they are solving some of the basic problems with the NLP models and NLP modelling.
The above-given examples of real-life use cases of the combination of ML and OR are some major examples that are consistent with the improvement. There can be various examples of this combination and also the only motive is to use the combination to improve the work strength and accuracy and benefit of the organizations.
In this article, we have seen what are the basics of operation research and how it can be combined with machine learning. The point to be noted here is that the machine learning models are related and concerned with the one task prediction whereas the operation research is concerned with the large collection of unique methods for specific classes of problems. As we have seen in the examples we can achieve higher accuracy and benefits using the combination of the ML and OR.