Oluwatosin Amosu
CEO
"At DataClear Consult, we're on a mission to revolutionize industries through data-driven solutions. We understand the unique needs of your business and offer cost-effective AI solutions. Let's embark on this journey together and set new standards in your industry."
PROJECT 1
Title: Customer Recommendation Analysis for an Airline Service Company Problem
Description: An airline service company wanted to improve customer satisfaction and understand the factors that determine whether customers will recommend their airline to others based on their flight experience. They sought insights into the key factors that influence customers' recommendations to help enhance their services and reputation.
Action: To address this challenge, a comprehensive data analytics and machine learning project was undertaken:
Data Collection: User reviews and feedback data were collected from the company's website through web scraping with python selenium. This dataset included customer reviews of their flight experiences, ratings, and whether or not they recommended the airline to others.
Data Preprocessing: The collected data was cleaned and preprocessed to handle missing values, remove duplicates, and perform text data preprocessing (e.g., text tokenization, sentiment analysis, and feature extraction).
Exploratory Data Analysis (EDA): Various statistical and visual techniques were applied to gain insights into the data. EDA helped identify trends, patterns, and correlations within the dataset.
Machine Learning Model: A predictive model, such as a classification model (e.g., LGBM, random forest, SVM ), was developed to predict whether a customer would recommend the airline based on the provided features. LGBM had the best f1 score of 95%. The model was trained, validated, and fine-tuned to achieve the best performance.
PROJECT 2
Title: Academic Paper Bulk Downloader for Researchers
Description: Researchers and academics often need to access and download multiple research papers from various platforms and sources. Manually downloading each paper can be time-consuming and inefficient. Researchers need a tool that allows them to automate the process of downloading academic papers in bulk from different platforms. They also prefer a user-friendly web interface for easy access and control.
Action: We developed an academic paper bulk downloader tool using Python, Selenium, and BeautifulSoup. The tool provides a web interface built withStreamlit, making it accessible and user-friendly. Here's how it works:
Platform Compatibility: The tool supports four academic paper platforms and repositories, IEEE, Science Direct, Link Springer, and Digital Library.
User-Friendly Interface: Users can access the tool via a web browser. They simply provide the URL of the page containing the list of papers they want to download.
Custom Folder Selection: Users can specify a destination folder on their local machine where the downloaded papers will be saved.
Batch Download: The tool scrapes the page for paper links, extracts the necessary information, and downloads the papers in bulk. It can handle downloading dozens or even hundreds of papers simultaneously.
Error Handling: The tool is designed to handle common issues that may arise during the scraping and downloading process, such as login requirements or CAPTCHA challenges.
PROJECT 3
Title: Deep Learning Meat Classification for Quality Assessment
Description: In the food industry, ensuring the quality and safety of raw meat products is of utmost importance. However, visually determining the state of raw meat, whether it's good, spoilt, or half-spoilt, can be challenging and subject to human error. To address this, there is a need for an automated system that can accurately classify the quality of raw meat based on visual characteristics.
Action: We developed a Deep Learning Meat Classification system using Python, PyTorch, and Streamlit. Here's how the project was structured:
Data Collection: We collected a comprehensive dataset of raw meat images, including various states such as good, spoilt, and half-spoilt. This dataset was carefully labeled to train the deep learning model. robustness.
Model Development: We built a convolutional neural network using PyTorch for image classification. The model was trained on the labeled dataset to learn the visual features associated with each state of raw meat.
Model Evaluation: We assessed the model's performance using metrics such as accuracy, precision, recall, and F1-score. Extensive testing and validation were conducted to ensure the model's accuracy in classifying meat quality.
Deployment with Streamlit: To make the model accessible to users, we created a user-friendly web interface using Streamlit. Users can upload images of raw meat, and the model will classify the quality as good, spoilt, or half-spoilt.
Feedback and Improvement: We continually collected user feedback and used it to fine-tune the model. This iterative process helped improve the model's accuracy and robustness.
PROJECT 4
Title:WhatsApp Chatbot for Bank Customer Service
Description: A bank aimed to enhance customer service by providing a convenient and accessible channel for customers to inquire about their accounts, transactions, and other banking services. They sought to deploy a chatbot on WhatsApp to efficiently handle customer queries, reduce response times, and improve overall customer satisfaction.
Action: To address this challenge, a WhatsApp Chatbot for Bank Customer Service was developed using a transformer-based model. Here's how the project was implemented:
Data Collection and Annotation: A dataset of customer queries and intents was collected and annotated. The dataset included various customer inquiries, such as balance inquiries, transaction history requests, fund transfers, and account-related questions. Each query was labeled with its corresponding intent.
Intent Recognition Model: A Universal Sentence Encoder model was trained to recognize customer intents. The model learned to classify incoming queries into predefined categories or intents.
Dialog Management: A dialog management system was implemented to handle multi-turn conversations with customers. The chatbot was designed to remember context and maintain coherent conversations, allowing customers to ask follow-up questions and receive relevant responses.
Integration with WhatsApp: The chatbot was integrated with WhatsApp using the Twilio API.Customers could initiate conversations with the bank's WhatsApp number and interact with the chatbot seamlessly.
Response Generation: For each recognized intent, the chatbot generated appropriate responses. Responses included account information, transaction details, helpful links, and instructions for certain transactions, among others.
PROJECT 5
Title:Plant Disease Recognition for Crop Health Monitoring
Description: Farmers face the constant challenge of protecting their crops from diseases that can significantly impact yield and quality. To address this issue, there is a need for an automated system that can accurately detect and classify plant diseases in real-time. This system should be user-friendly and accessible to farmers for early disease identification and intervention.
Action: To provide an effective solution, a Plant Disease Recognition system was developed using pyTorch and deployed as a web application using Streamlit.Here's how the project was structured:
Data Collection: A comprehensive dataset of images depicting various plant diseases affecting crops like maize, tomato, potato, raspberry, and others was assembled. Each image was labeled with the corresponding disease category.
Model Development: A deep convolutional neural network (CNN) model was built using PyTorch. The model was trained on the labeled dataset to learn the visual characteristics and patterns associated with each disease.
Model Evaluation: The model's performance was assessed using classification metrics such as accuracy, precision, recall, and F1-score. Robust testing and validation were conducted to ensure reliable disease detection.
Deployment with Streamlit: To make the model accessible to farmers and stakeholders, a user-friendly web interface was created using Streamlit. Users can upload images of plant leaves or crops, and the model will classify them into one of the 38 disease categories.
Feedback and Improvement:Continuous feedback from users and additional data collection allowed for model refinement and performance improvement.