Introduction

The report points out AI-based solutions to the following three business problems: forecasting the value of houses, waste classification through pictures, and gold price forecasting. Every task is peculiar for the dataset and machine learning model to solve decision-making tasks such as real estate valuation, waste management, and financial forecasting.

Task 1: Real Estate Analytics with Tabular Data

Problem Statement: The problem is a prediction problem, where house prices are to be estimated from a dataset containing various features of houses sold in a city. The major business problem is constructing AI models that accurately estimate house prices using features such as bedrooms, bathrooms, square footage, and other variables.

Data Preprocessing: Cleaning the dataset for missing values and unnecessary features, 'id', 'date', and 'zipcode' were removed to make the dataset relevant. Variables such as price, bedrooms, and bathrooms were outlier-capped to maintain model stability. Feature engineering was conducted and new variables were created such as features including price per square foot and age of property, which enhances the robustness of the model.

AI Model Development: The developed work implemented a Linear Regression model and a Multi-Layer Perceptron (MLP) Regressor, comparing performances (Khosravi, et al., 2022). In the case of MLP, tuning via grid search was done to find the best configuration concerning the number of hidden layers, iterations, and regularization parameters. It split the dataset into 70% for training and 30% for testing, standardizing all its features to enhance model performances.

Results and Analysis: The R?2; value for Linear Regression is 0.91, while the initial MLP Regressor outperformed this result with a score of 0.95. After tuning, the MLP further reached an R?2; close to perfection at 0.999, though subsequent optimization managed to degrade the performance slightly. These results showed that the tuned MLP Regressor provided the maximum predictive accuracy, and thus it is more appropriate for the task.

The best MLP model performed extremely well because of its capability to capture complex patterns. For real-world implementation, further testing on unseen data and performance monitoring will be required. Challenges to the scaling of the solution effectively include variability in data and computational costs.

Task 2: Waste Classification with Image Data

Problem Definition: Waste management requires an efficient sorting method for recycling. The dataset consists of 2,864 images, divided into six classes of waste: cardboard, glass, metal, paper, plastic, and vegetation. There is a need to develop machine learning models particularly CNNs among others to classify these waste types accordingly.

Data Preprocessing and Exploration: The imbalance in the classes of the dataset is improved using augmentation techniques: rotation, shift, and zoom. Images were resized to 224 x 224 considering input for models such as VGG16. In all categories intensities and sizes of pixels are identical.

AI Model Development: Three CNN architectures were explored:

• Baseline CNN: A simple model using three convolutional layers combined with max-pooling and dropout (Nnamoko, Barrowclough, and Procter, 2022).
• Deeper CNN: A slightly larger model, including five convolutional layers with more regularisation added.
• VGG16 Transfer Learning: It includes pre-trained VGG16 with custom classification layers on top.

Each model used categorical cross-entropy loss and was adjusted for class weighting due to the class imbalance.

Results and Analysis: The baseline CNN achieved a validation accuracy of 53.4%, while that of the deeper CNN was 55.9%, and for VGG16 Transfer Learning was 57.3%. The worst-performing class was metal, likely due to its reflective texture being similar to plastic and glass.

Future work may consider improvements: gather more diverse data, develop better preprocessing techniques, and fine-tune model architecture. Domain-specific knowledge could also help enhance classification performance. The idea of such optimizations will help real-world deployments of these models for the variability of real-world images of wastes and will enhance overall classification performance across all categories of wastes.

Task 3: Gold Price Forecasting with Time-series Data

Problem Definition: The task to be accomplished now is to forecast the Gold Price in USD for 2 weeks using multivariate time series data. This data typically runs from 1985 to 2023. It encompasses daily gold prices in USD, EUR, GBP, INR, AED, and CNY. The best forecast will depict an optimum trading strategy in light of minimum risk in these highly volatile markets (Chimmula, and Zhang, 2020).

Data Pre-processing: The preprocessing steps included the preparation of data by handling missing values, normalization using the MinMaxScaler technique, and splitting into training and test sets correspondingly (pre-2022) and (2022-2023). Series were generated using lagged values of 14 days. Prices in several currencies helped to show global trends properly.

AI Model Development: The development and tuning of two RNN models LSTMs and GRUs were made for the forecastings by using different hyper-parameters-space with varied units of 50, 100, and 200, dropout rates of 0.2 and 0.3, and epochs of 20 and 30. Later on, the training of models was based on multivariate input to predict gold prices in USD.

Results and Analysis: Comparing the performances of many models, the best performance was that of the GRU with 50 units and 0.2 dropouts. This has an MSE of 296.92, RMSE of 17.23, and 13.16 of MAE. GRU was observed to develop the finest performance since it captures sequences quite effectively when preventing overfitting.

This should further improve in a real-world application deployment with more feature engineering on the data, tuning of the hyperparameters, and periodic retraining of the model to adapt to the market fluctuation better. Adding more market-specific variables can potentially further add to the prediction accuracy.

Conclusion and Recommendations

The report provides the best AI solutions for estate, waste, and gold price prediction. The best models are MLP for predicting house prices, VGG16 for waste classification, and GRU for forecasting the price of gold. More data collection, model refinement, and testing continuously will scale up the performance in each of the cases for real-world deployment.

Are you struggling to keep up with the demands of your academic journey? Don't worry, we've got your back!
Exam Question Bank is your trusted partner in achieving academic excellence for all kind of technical and non-technical subjects. Our comprehensive range of academic services is designed to cater to students at every level. Whether you're a high school student, a college undergraduate, or pursuing advanced studies, we have the expertise and resources to support you.

To connect with expert and ask your query click here Exam Question Bank