CT118-3-3 ODL Optimisation and Deep Learning Individual Assignment | APU

This Assignment Has Two Parts: 

Assignment 1 covers Data Preparation and Preprocessing, and the justification of Neural Network algorithms for data analytic applications in real-world problems. 

Assignment 2 covers applying deep learning methods with optimisation concepts for practical solutions. 

Objective: The main objective of this assignment is to find an optimal deep learning model for a problem supported by historical data. This is an end-to-end project that addresses optimality in hyperparameter and model parameter selection using optimisation procedures while building suitable deep learning models.

Assignment 1 (40 marks) 

Task 1: Dataset & Algorithm selection (20 Marks) 

Choose a suitable medium/large size secondary dataset from the suitable and possible open data resources. Critically analyse the dataset and suggest two predictive neural network (deep learning) models that may be suitable for your dataset, and discover the optimised solutions for the defined problem. Produce a comprehensive literature review on the past research work on the same or a similar chosen problem. Justify your selection of such a dataset on the suitability for this assignment.

Task 2: Data Preparation (20 Marks) 

Perform an Exploratory Data Analysis (EDA) on your dataset using the suitable tools and techniques. All the data preparation activities are expected to take place in detail and be clearly reported.

Assignment 2 (60 Marks)

Task 1: Model Building (30 Marks)

In this part of the assignment, build the deep learning model based on the selection suggested in Assignment 1 - Task 1. You are recommended to select different deep learning algorithms to work on. Conduct basic model initialisation, training, and evaluation. Choose a suitable training algorithm, evaluation metrics and build your basic model.

Task 2: Model Tuning (10 Marks)

The model hyperparameters shall be tuned and validated. List and explain the Meta / Hyper parameters of your model. Choose the optimal Meta parameters of your model using suitable searching techniques. You may use either grid search, random search or Keras Tuner methods. Design and fine-tune the model to build the final model.

Task 3: Model Evaluation & Discussion (10 Marks) 

Select evaluation metrics and briefly describe them. Perform evaluation and present your results.

Task 4: Conclusion (10 Marks)

Produce a critical comparative analysis with the peer models, including the comparison of the predictive models reviewed in the literature. Suggest how your results may be improved further.