Abstract:
"
With the increase of vehicle imports, used vehicle market has become popular and interesting.
With the market popularity and no prior knowledge about used vehicles, many people get fraud
to the vehicle dealers and vehicle buyers. Therefore, this becomes an important and interesting
problem. In my research, I proposed a system to predict used vehicle price for the Sri Lankan
vehicle market. Therefore, I decided to implement a web application that can predict the price
of used vehicles. As first priority, I collected the dataset from Ikman.lk, which is a website that
can sell and buy used vehicles. I was able to collect 21510 rows and after several phases of
data cleaning process there was 17677 rows. In the data preprocessing phase I have used label
encoding as well as one hot encoding for the categorical variables. Before selecting an
appropriate algorithm for the model, I have used three algorithms and did a comparative study
on each other performance. XGBoost algorithm performed better than linear regression and
random forest. XGBoost training and testing results were 97.14% and 91.72%. The best performed algorithm optimized with Bayesian optimization and compared the result. Then the
web application built with flask using the prediction model which gave the best result."
