NLP Sentiment Classification -Classifies text sentiment with 91% accuracy

The deep learning model uses natural language processing (NLP) to classify positive or negative statements written about four different subjects: education, finance, politics, and sports. It demonstrates practical steps in preprocessing, modeling, and evaluating performance, serving as a useful guide for comparable NLP projects. This type of AI-powered solution can be utilized for a variety of tasks, from sorting customer reviews and social media posts, to analyzing healthcare patient data and detecting fraud.

After merging the datasets, I generated synthetic statements to balance positive and negative samples, enhance sample representation, and improve the model's accuracy.

I evaluated multiple different models and selected a neural network (a type of machine learning model inspired by the structure and functioning of the human brain) optimized with 91% accuracy.

The confusion matrix visualizes the model's excellent accuracy (91.46%) and confirms it performs well in predicting both positive and negative cases.

I tested the model on several different statements to confirm the model's efficacy and made a function to test the sentiment on custom inputs. The model accurately classified the test and custom statements.

This sentiment classification deep learning model demonstrates the power of machine learning and NLP techniques in accurately understanding text-based emotions. With an overall accuracy of 91%, the Neural Network model effectively balances precision and recall across both positive and negative classes. This model has a wealth of practical applications, such as customer feedback analysis or sentiment tracking in social media, offering a solid foundation for further development and implementation.

View the project: Natural Language Processing Sentiment Classification

Machine Learning for Crop Yields - Predicts yields with 98% accuracy

For this project, I looked at over 20 years of crop yield data from the Food and Agriculture Organization. I then used machine learning techniques to make predictions for future yields. This machine learning analysis was completed using Python. This project could have practical applications for farmers who want to predict yields based on environmental factors. This analysis workflow could be catered to specific regions, crops, farms, or different environmental conditions.

We can see here that annual yields increased over time.

Pesticide use also increased steadily with yields.

The total amount of fields increased gradually over the 24 year period.

This correlation heatmap shows the correlations between the different numerical variables.

Next, I tested different machine learning models for yield predictions, including a neural network model. After testing several models, I selected Random Forest Regressor for the final model, with an R-squared score of 0.97. These charts show the predicted values vs the actual values. I used Cross-Validation RMSE to help check the model for overfitting. I optimized the final model to ensure overfitting was not an issue, ensuring it's efficacy in making real-world predictions.

The final model achieves an impressive R-squared score, averaging 0.98 across the training and test sets (0.97 for the training set and 0.99 for the test set). The cross-validation RMSE and test RMSE are also very similar (13745.45 for the test RMSE and 13893.52 for the cross-validation RMSE), indicating the test performance is representative of the model's ability, and overfitting is not an issue. While there is a slight difference in the training RMSE and the test RMSE (7901.43 for the training and 13745.45 for the test), this difference is reasonable given the different sizes of the datasets. The MAE values are relatively close as well (3223.64 for the training and 5688.32 for the test), suggesting the model does not overfit the training data. The scores prove the model's efficacy, it does an excellent job of predicting the yields, accounting for over 98% of the variance in the target variable (yield). This machine learning workflow could serve as a template for predicting agricultural yields based on historical trends and environmental factors benefiting farmers, policymakers, and agri-tech innovators.

View the Project: Crop Yield Analysis and Predictions

Clothing & Accessories Sales Trends - Analysis and Customer Segmentation

In this exploratory data analysis, I examined customer trends in clothing and accessories sales across the United States. The project visualizes various customer trends and patterns. This exploratory analysis was conducted with Python. The accompanying dashboard was made with Tableau. From understanding sales trends to developing advanced customer profiles and procuring insights into customer behaviors, this type of analysis can be incredibly valuable for retail enterprises, startups supplying consumer goods, and medium-sized clothing and apparel companies looking to utilize AI-driven outreach.

I utilized various methods to test for significance, including ANOVA and the Chi-Square test. The most notable associations identified were between promo code usage and subscription status, gender and subscription status, and season and purchase amount.

There was a clear pattern indicating that women were not subscribers, with none of them being subscribed or using promo codes. This should be a primary focus for marketing efforts: understanding why women are not subscribing and offering targeted promo codes that include free or discounted subscription options to encourage engagement.

I used k-means clustering to create three distinct groups of customers. The clusters were divided into low spenders who gave moderate ratings, high spenders who gave excellent ratings, and moderate spenders who gave low ratings. Personalized emails or advertisements could be tailored and sent to customers based on their respective clusters. For example, high spenders who gave excellent ratings could receive a discount promo email expressing gratitude for their loyalty. Low spenders who gave moderate ratings could be offered competitive discounts to encourage higher engagement. Moderate spenders who gave low ratings could receive an email with an exclusive offer to improve their satisfaction and incentivize future purchases. This targeted approach ensures more effective and customer-centric marketing.

Sales were greatest in the fall and winter, though demand for different items varied. As mentioned, the data indicates that women did not use promo codes and were not subscribers. With women comprising close to a third of the customers, something should be done to address this trend. Customers could also be sent emails based on which cluster group they were placed into. I have included an interactive dashboard visualizing the sales trends. The dashboard was made using Tableau. To view other dashboards I have made, view the dashboards page. This type of exploratory data analysis on clothing and accessories trends can uncover patterns in sales and customer behaviors, providing actionable insights for brands, retailers, and consumers.

View the Project: Clothing and Accessories Analysis

Analysis of Avocado Sales -Analyzes sales trends using SQL

In this analysis, I examined avocado sales and prices from 2015 to 2021. This data comes from the Hass Avocado Board website. The analysis was performed using SQL.

Here I examined correlations between the different avocado sizes and the number of bags sold.

The strongest correlations were between the number of total bags sold and the number of small, medium, and large avocados sold. I also looked at the top 50th percentile and top 90th percentile for total sales. Click below to view the full SQL analysis. This type of project could be useful for understanding trends in avocado sales, like pricing, demand, and regional preferences. It can help retailers set better prices, improve supply chains, and predict demand more accurately.

View the Project: Avocados Sales Analysis with SQL

Spotify User Data Analysis -Explores Spotify listening trends

I performed an analysis of Spotify user data to examine trends and insights. The data was collected anonymously using Google Forms. This project looks for trends in the user data, such as user demographics and behaviors. The analysis was completed using RStudio.

Spotify users tended to be younger, with the largest age group being users age 20 to 35. Most users had used Spotify for a year or more. Females responded to the survey far more than other genders.

Spotify users primarily listened to music, but podcasts were also popular. Users reported melody as the most preferred musical genre.

Recordings had a mean rating of 3.5 out of 5 stars. Users listened the most while traveling or during their leisure time. The night was the most reported time slot for users to listen to music. Comedy was the most preferred podcast genre for users who indicated a favorite. Most users listened to music for relaxation and stress relief. Below is a Microsoft Power BI dashboard I created to help visualize the data. This type of project could be useful for understanding listening habits and trends among Spotify users. It can help identify popular genres, artists, or times of day for music consumption, which could guide personalized recommendations or marketing strategies. Artists and streaming platforms could also use the insights to tailor content and improve user engagement.

View the Project: Spotify User Data Analysis

Predicting Sales Based on Budgets -Predicts sales with 96% accuracy

In this project, I utilize machine learning to forecast total sales based on marketing budgets. The analysis was performed using Python. I began by cleansing and formatting the data. Then I made a correlation heatmap and scatter matrix to visualize the strongest correlations within the dataset.

I tested several models and found that a polynomial regression model implemented using a pipeline performed the best.

The model is performing well with an average r-squared score of 0.9655. With only minor differences between the training, validation, and test sets, the model is not overfitting. The cross-validation RMSE is close to the training and test RMSE, again confirming the model is not overfitting. These scores confirm the final model's efficacy, it predicts an average of 96.6% of the variance in sales based on marketing budgets. This project could be useful for businesses looking to understand how their marketing budgets impact sales performance. It can help identify which marketing channels provide the best return on investment and guide future spending decisions. Companies can use these insights to optimize their campaigns and boost overall sales.

View the full project: Predicting Sales Based on Marketing Budgets

Home

Portfolio Continued

Dashboards

Web Applications

Contact

Page updated

Google Sites

Report abuse