Colorimetry is the technique used to determine the concentration of colored compounds in a solution. The aim of the project includes developing machine learning models and algorithms, to predict colorimetric property such as pH of an analyte using primary colors(RGB) values. Using the ML model which predicts pH value, we can obtain real pH predicted value using a web application.
Project presentation is available at https://github.com/bahl24/pH-data-prediction/blob/master/Poster%20-%20pH%20predictor.pdf
The dataset is in CSV(comma separated values) format obtained from https://www.kaggle.com/robjan/ph-recognition . The data consists of 4 columns having RGB color values. Three columns have the features - blue, green, red and the fourth column has the label which is the pH(0-14). There are 653 distinct data points(rows) covering all the pH range from 0-14 with almost equal distribution.
I compared the performance of 6 common ML algorithms:- Linear Regression, Logistic Regression, KNN, SVM, Decision Trees & Random Forest Classifier, RFC gave the least MSE (0.51) and highest accuracy (72%). Thus, I trained and built an RFC based ML model and saved it, to use it in the web app.
Pickle is the standard way of serializing objects in Python. Pickle operation can be used to serialize ML algorithms and save the serialized format to a file. Pickle can save this as a binary file, which can be loaded in another program to deserialize and make predictions with the model.
Next step after saving the ML model, is to develop a web application to enable users to actually use the model and make the predictions. I've used VueJS (a JS framework) to build the frontend part of the web app.
As the pickle model is Python-based, we need to develop a Flask based REST API backend to fetch and post data via HTTP to the web page.