Startup Ideas Tips
How to Find an Interesting Idea for Your Next Machine Learning Project
A three-step process to increase the amount and quality of ideas for your next side project
I have wanted to learn machine learning for quite some time now. As many people do when they want to learn something new I did multiple courses, scrolled through Youtube for explanatory videos, and read about different elements of machine learning on the internet. Although all of that gave me a great overview and a basic understanding of what machine learning is, I realized that I didn’t acquire the skills necessary to make something useful by myself.
I needed to get hands-on experience with relevant projects to make the newly acquired knowledge useful. I could have tried to do a competition on Kaggle or found a clean dataset to work with, but I didn’t want to do something someone else had done before. I believe that you learn most when you try to challenge yourself to accomplish something no one else has done. Therefore, I developed a system of my own to get the ideas flowing.
To use the system presented above, choose an industry that you are familiar with and break down the value chain of that industry. The more steps you are able to break the value chain into, the better. Above, I have broken down agriculture into production, harvesting and transport, processing, distribution and packaging, and wholesale and retail.
After you have written down the value chain, you choose the technology you want to use or learn during the project and break that technology into as many parts as possible. Machine learning can be broken down into supervised learning, unsupervised learning, and recommender systems (which can be both supervised and unsupervised). Supervised learning encompasses both regression and classification tasks, while two famous examples of unsupervised learning are clustering and anomaly detection.
With the header column and header row set, it’s time to fill in the cells with ideas for new machine learning projects. Go through the table column for column, and fill in all the cells with new ideas solving the needs at different steps in the value chain. The beauty with this system is that while filling out the cells, you will be able to see how a technology that can be used in one of the steps in the value chain can be used in a slightly different way in a different step. Hence, you will be able to discover more ideas for new projects.
The beauty with this system is that while filling out the cells, you will be able to see how a technology that can be used in one of the steps in the value chain can be used in a slightly different way in a different step.
One of the more interesting ideas from the table above is the one in the intersection between classification and distribution and packaging: classifying vegetables and fruits into different classes based on size, using computer vision. For some types of vegetables, there are requirements for the size of the vegetables that are sold. Many farmers, therefore, use a lot of time to manually sort out vegetables that are not inside the right size range. However, this task can easily be done by machine learning, and to build a prototype the only things needed are a camera, a Raspberry Pi, an inclined surface for the vegetables to roll slowly down, and something that can be moved by the Raspberry Pi so that the vegetables will fall into the right basket after being classified by the computer vision algorithm.
Although the system is designed specifically to discover more ideas for side projects, it can also be used to come up with great ideas for B2B businesses. However, I will advise you to focus more on problems and secrets rather than technology subsets in order to discover truly important ideas. A more appropriate system will have the value chain in the header column, and then have columns for problems and secrets in the header row. For a detailed description of how this can be done for B2C businesses take a look at the below article.
Personally, I am extremely interested in the venture capital and startup world and wanted to learn more about that while developing my machine learning skills. Therefore, I constructed a similar system as the one above but switched out the agriculture value chain with the venture capital value chain. After filling out the entire table, I realized that I wanted to build a startup success predictor using machine learning. I am now in the closing stages of that project, and have learned a lot about all the important steps of a typical machine learning project: identifying relevant data, collecting data, explorative data analysis, data cleaning, choosing/developing a model, training the model, and evaluating it. I probably wouldn’t have been able to come up with that idea if I hadn’t used the system described above.
Readers are encouraged to take the system presented in this article and try it for themselves, by following the three steps below.
- Break down the value chain of an industry to focus on a specific need at a time
- Break down the technology you would like to learn more about into sub-technologies
- Fill in ideas for how to use the different sub-technologies to solve needs in different steps of the value chain in the chosen industry
