3
$\begingroup$

Let's say I have a database of freight orders. The job is to match freight carriers with customers who need their freight moved. I have the customer's information, the freight carrier's information, and all details related to the freight orders including date ordered, date shipped, the amount of money it took to hire the freight carrier, and whether a carrier was even found to ship the order.

If I have thousands of these past freight orders, could I use machine learning to look at future freight orders to predict whether or not a freight carrier will be found to move it?

Bonus: If it is possible, what steps would I need to take to find the best data points to focus on? From what I understand, I need to convert everything to a number in order to train the classifier, but I am having trouble figuring out what data features are going to help make these types of predictions.

I have been studying how to do machine learning and I am not looking for somebody to tell me everything there is to know on the subject, I just don't know how to determine what data points are going to be useful and am also looking for an answer to whether or not this is something machine learning can do(or if it's something a beginner in machine learning can do). Sorry if the question seems vague, it's kind of hard to articulate on a subject you are just starting to learn about. If anybody has materials they can link that would help me to better understand these things, I would appreciate that as well..

$\endgroup$
7
  • 3
    $\begingroup$ This sounds too broad to me. Community votes? (Your first step is to study how machine learning works. It sounds to me like you are asking "I don't know anything about machine learning; can you tell me everything I need to know about the field to solve this problem?" That sounds too broad to me. I'd suggest you start by doing some self-study, then come back when you can formulate a more specific question. Alternatively, can you edit the question to ask about a specific aspect of the problem?) $\endgroup$
    – D.W.
    Sep 16, 2015 at 15:45
  • $\begingroup$ Well I have spent hours studying machine learning, and there is very little information I have found that is helpful due to not knowing much about the subject, so finding places to start is difficult to say the least. I know how to use the library I am using to do what I need to do, the problem is that I don't know much about statistics and I don't know what data points are going to give me the best predictions. The rest of it I can figure out. I don't need any help with the library. $\endgroup$ Sep 16, 2015 at 16:18
  • $\begingroup$ @DavidRicherby, The official question is, "Can I use Machine Learning to Solve this Problem?". Listing steps I would need to take to get good predictions would just be icing on the cake, but I think the question is definitely on-topic. I guess the library wouldn't matter honestly, I am more concerned with if machine learning can solve the problem or not. $\endgroup$ Sep 16, 2015 at 16:19
  • $\begingroup$ @DuckPuncher I agree that the edited question is wholly on-topic. $\endgroup$ Sep 16, 2015 at 16:51
  • 1
    $\begingroup$ "What data features are going to be useful?" are not computer science questions. Those are questions that relate to your particular application domain: freight ordering. So, I don't see how we could possibly tell answer that for you -- you'll have to tell us. But in a similar vein: Have you read about feature selection? $\endgroup$
    – D.W.
    Sep 16, 2015 at 18:18

2 Answers 2

4
$\begingroup$

Here's the thing to know: machine learning assumes that there is some sort of statistical distribution, and you "learn" that distribution, in order to get the probability of some event.

The common saying in machine learning is "garbage in, garbage out." You have a bunch of random variables, but if those variables are all statistically independent, then you won't get anything useful from machine learning.

Say, for example, that there was a strong correlation between the amount of money paid for the carrier, and whether the item is shipped. Machine learning would likely be able to discover this relationship.

Or, if there were certain periods of the year where an item was more or less likely to be shipped. Machine learning could find this out.

But, if there's no underlying pattern in the data you give to your training algorithm, then you will get no useful information out of it.

For your question of finding the best data points: don't. That's what the machine learning algorithm is for. You give it your data, and it looks at which ones are statistically relevant, given some threshold. The whole point of machine learning is letting the algorithm do that for you, and more importantly, that's all the algorithm can really do for you.

Generally, these algorithms will work better when they have more data, so don't go out of your way to remove what data you're giving it.

And, remember, that ALL these algorithms are doing is statistics. If it says that the item is shipped with probability 0.6, don't be surprised when it isn't shipped. And, it's possible that it will say the item is shipped with probability 0.99, but it won't ship, because there's some variable which is hugely important in real life, which you didn't have recorded for your data set. If that variable isn't correlated with your data, then your model will be no good.

$\endgroup$
4
  • $\begingroup$ this is exactly what I was looking for! I have troves of data and we think there are patterns at play here. Thanks! $\endgroup$ Sep 16, 2015 at 18:30
  • $\begingroup$ You might want to look at WEKA as a toolkit for doing your tasks. I'm guessing you'll want to start with decision trees, SVMs, and Naive Bayes as the simplest classifying algorithms. But that's not more StackOverflow territory than CS.SE. $\endgroup$ Sep 16, 2015 at 18:34
  • $\begingroup$ thanks. I am actually using scikit-learn along with a few other python modules to tackle this, but if it doesn't seem to do the job I need it to do, I will give it a shot. I'm willing to try anything at this point. Thanks again! $\endgroup$ Sep 16, 2015 at 18:43
  • $\begingroup$ Yeah, Python has a library for everthing, so there's a good chance scikit-learn will do the trick for you. Good luck! $\endgroup$ Sep 16, 2015 at 18:45
0
$\begingroup$

Yes, you can definitely use machine learning to predict the likelihood of an order being shipped. This prediction can be helpful for various purposes, such as optimizing inventory management, improving logistics, and providing better customer service by anticipating delays.

Here's an example of how you could approach building a machine learning model to predict the likelihood of an order being shipped on time:

Data Collection: Gather historical data on orders from your database or e-commerce platform. The data should include features such as order date, shipping date, order location, shipping location, product type, shipping method, and any other relevant information.

Data Preprocessing: Clean and preprocess the data to handle missing values, encode categorical variables, and scale numerical features if necessary. This step is crucial for ensuring the data is in a suitable format for machine learning algorithms.

Feature Engineering: Create additional features that might be useful for predicting shipping likelihood. For example, you could calculate the time difference between order date and shipping date, or extract features related to the shipping destination, such as distance from the warehouse or average shipping time to that location.

Model Selection: Choose an appropriate machine learning algorithm for the prediction task. Some common choices for binary classification tasks like this include logistic regression, support vector machines, decision trees, random forests, or gradient boosting algorithms.

Model Training: Split your preprocessed data into training and testing sets. Train the selected machine learning model on the training set using the labeled data (orders that were shipped on time and those that were not).

Model Evaluation: Evaluate the performance of your model on the testing set using metrics such as accuracy, precision, recall, or F1 score. These metrics will help you assess how well the model is predicting the likelihood of orders being shipped on time.

Model Deployment: Once you're satisfied with the model's performance, deploy it to make predictions on new orders in real-time or on a periodic basis.

Example: Let's say you work for an online retail company, and you want to predict the likelihood of an order being shipped on time. You gather historical data on orders placed over the past year, including order date, shipping date, order location, shipping location, and product type.

After preprocessing the data and engineering features like the time difference between order date and shipping date, you decide to use a random forest classifier for the prediction task. You split the data into a training set (70%) and a testing set (30%) and train the model on the training set.

After training, you evaluate the model on the testing set and find that it achieves an accuracy of 87%, which is promising. Now, you can deploy this model to predict the likelihood of shipping orders on time for new orders as they come in, helping your company optimize logistics and improve customer satisfaction.

Keep in mind that the success of the model depends on the quality of the data and the features used, so continuous monitoring and refinement of the model may be necessary to maintain accurate predictions over time.

I learned to do this with the help of https://www.eduonix.com/live-data-science-certification-program

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.