Deep-Learning (Modern Machine Learning)

Deep Artificial Neural Network for predicting student performance UCI Repo Dataset. The experimental work in this repo intends to extend on my previous Master's research on Machine Learning Evaluating AdaBoost, 2015 for predecting student performance. This time, let's do it the Deep-Learning way! ¯\_(ツ)_/¯

Model Accuracy 🤗

With a very small dataset the aspect of accuracy can be concerning. However, Currently the model is "close to accurate", but can be further improved with additional techniques such as parameter-tunning, cleaning out outliers from the dataset, fitting the right number of hidden layers to the model/classifier etc...

The training accuracy was 100%, but the test accuracy was 85%... Now, one would wonder what happened? 🤔 Why such a 15% difference/discrepancy in accuracy? 🤔 Small Dataset?🤔 clean out outliers?🤔 too many features(independent variables)?🤔 Needs further Parameter Tuning? 🤔 K-Fold Cross-Validation?🤔 Too many/few Hidden Layers? 🤔 Overfitting? 🤔 Underfitting? 🤔 This is exactly what makes Deep Neural Nets fun right? 🤣

Training Accuracy

Training Loss

Part-1 of 2

Part-1 & Part-2 are exactly the same except that Part-1 utilises high-level abstraction frameworks such as a combination of Keras & Tensorflow. Part-2 involves lower-level abstraction frameworks such as a combination of Numpy & Tensorflow.

Part-2 dives deeper into most of the interesting mathmatics that happen behind the scenes, whereas in Part-1 most of the mathematics is abstracted away from the user. Oooh 😱, there's no fun in that, right!😱? No Mathematics 😵 !!!

We love Mathematics 😎... But there's a lot value in modern high level-abstraction Machine Leanring techniques for beginners & newbies.

My vision is to inspire more Software-Engineers to co-create A.I driven applications.

Have Fun ✌(-‿-)✌