5 Unique Final Year Machine Learning Project Ideas
Nope! Nothing like sales price prediction here
I present to you what I like to call my idea dump. But believe me when I say it’s not your usual idea dump.
In this blog, I’m going to (as briefly as possible) talk about some interesting machine-learning projects that I’ve randomly thought about in the past but might never actually take on.
I mean why waste a good (AND FUN) idea, right?
Hopefully, you’ll find my ideas useful (or at least inspiring) and well, bring them to life as your final year project.
1. Cloth and Accessories Matching Analyzer
I’m bad at naming things but that aside, I need this in my life! SERIOUSLY.
The idea here is quite simple and straightforward: You train the model such that the users can feed it with photos of themselves dressed in an outfit and the model tells them if the outfit rocks or not.
It would be nice if the model were a bit more specific as to what the problem with the outfit really is. Are the shoes off? Do the colors not match?
It would also be useful if the model points out what really makes the outfit stand out if it’s a good one. Do the colors go well together? Does it just fit the specific body shape?
So yeah, simple idea that would save lots of lives ><
Some libraries/ frameworks you may need include OpenCV, TensorFlow, Scikit-learn, and Streamlit.
Datasets that may be useful: Fashion Datasets with Outfit Ratings
2. The Good AI Skin Doctor
This is a combo of two ideas that have been on my mind for some time now.
One is the facial skin care routine recommendation system and the other is the skin disorder analyzer.
The facial skin care recommendation system basically asks you a couple of questions about the nature of your skin and perhaps requests some images too.
And then recommends the most suitable skin care package tailored to your specific type of skin (that is it’s more guaranteed to work well with your skin) and budget.
Alternatively, it could recommend natural solutions based on your input data.
The second model which is the skin disorder analyzer does something similar: asks questions about some skin conditions you may have noticed (recently) i.e. discoloration, rashes, and so on.
Then it tells you what precise skin disorders/diseases you’re experiencing as well as recommends suitable treatment and therapy.
Some libraries/ frameworks you may need include OpenCV, TensorFlow, Scikit-learn, and Flask.
Datasets that may be useful: Skin defect data, facial skin data, questionnaire data
3. A Text-to-Speech Model Dedicated to Studying
When I feel too tired to study, I often consult my text-to-speech app Natural Reader to read me my course materials (mostly slides and PDFs).
But I find myself zoning out most times and I also find it hard to keep up with the AI’s monotonous voice.
Now this solution is not about the voice (the next one is but let’s focus on the real deal for now).
The solution is to develop a more complex system where the AI not only reads my PDFs to me but also explains each paragraph as it progresses.
We can take it a notch further, and let the AI be able to discern which paragraphs are already self-explanatory so as to not waste time over-explaining basic/ already explained concepts.
Some Libraries/ frameworks that may help carry out this project:
- TensorFlow
- NLTK (Natural Language Toolkit)
- gTTS (Google Text-to-Speech)
- Flask(For deployment)
The datasets you may need:
- Dataset of study materials in various formats, such as PDFs, slides, textbooks, etc.
- Labeled data where each paragraph is annotated with an explanation or difficulty level.
- ASR Data (audio data and corresponding text transcriptions)
4. A TTS -STT Celebrity ASMR Bot
This idea is literally the most random thing ever. But hear me out.
We all love our Snapchat AI Chatbot, It tells us what we want to hear (well, sometimes) and it’s always so gentle and polite.
And then also, at certain points in our lives, we may have imagined what it’d be like to talk over coffee with the fine beings we see on TV.
So how crazy would it be to have Chatbots, actually audio bots that can imitate these celebrities and have vocal conversations with us, like the talk over coffee but without the coffee?
I imagined that this idea would be able to come to life if someone built a special chatbot that does more than just ‘chat’.
First, it turns the text conversations into ASMR audio (not the emotionless voices we mostly have around). But it doesn’t stop there, it also makes the audio sound like famous celebrities.
Second, it ‘listens’ to what we say and changes it into text so that it can process our words and give us that satisfying ASMR vis-a-vis celebrity voice response.
So, it’s like casually chatting with your Snapchat AI bot, but this time it’s an oral conversation; you’re talking to it and it’s responding as fast as it can. And the best parts: the ASMR and the fact that it sounds like your favorite celebrity.
Libraries/Frameworks: TensorFlow, gTTS, NLTK , Flask
Datasets:
- Celebrity Voice Data
- ASMR Audio Data
A less interesting version of this idea would be an improved TTS system where the AI voices sound more like professional speakers who understand the importance of tone, pitch, and so on, instead of the dry AI voices we mostly see around.
5. Future Spouse Compatibility Analyzer/ Adviser
I got this idea in the most hilarious way possible and I’d love to talk about it but that’s too long a story and perhaps out of the scope of this blog post.
So I’ll just cut to the chase.
During a talking stage, or courtship with someone we prospect to be our life partner, we tend to ask them lots of questions.
Because we want to know as much as we can about them and discern if they’d be good life partners.
So one day I thought, the answers they give us are actually a form of data that can be used to logically (let’s face it, people don’t really think logically when they start getting emotionally connected with someone) decide if someone’s going to be a good fit for us and also predict potential conflicts that may arise in the future based on our personality clashes.
Isn’t that super cool? Having a model do all this logical thinking and analysis for you while you keep getting more data (spending time with the potential life partner) to improve its performance.
Okay, that may have sounded a bit sus but it is what it is.
I imagine this model to be very flexible and to require high dimensionality data.
And maybe when it has sufficient data on your prospective partner, it will be able to give relationship advice (that’s not based on sentiments as our friends do at times) when things go South.
Libraries/ frameworks: TensorFlow, NLTK, Scikitlearn, Streamlit.
Datasets to get: questionnaire data, relationship outcomes data
Conclusion/ Call-to-Action
Whew! So glad I got all of that out finally. Even if you don’t adopt any of my ideas for your final year project, I hope you’ve gotten some inspiration for your next ML project.
If you enjoyed this article, please leave as many claps (up to 50) and share with any final-year students who may need to see this.
And if you’re feeling more generous, feel free to buy me a coffee by clicking the button below:
Also, do let me know in the comments if you have any other ML project ideas that never made it out of your head ><
Bye for now!
