Browsing by Author "Ibrahim ODETUNDE"

Now showing 1 - 1 of 1
  • No Thumbnail Available
    Item
    Comparative Analysis of Selected Classification Algorithms on Weather Data
    (Faculty of Information and Communication Technology, Kwara State University, Malete., 2023) Shakirat Ronke YUSUFF; Ibrahim ODETUNDE; Jumoke Falilat AJAO
    Numerous human daily activities, such as transportation, farming, and commerce, are weather-dependent. Scientists have used a range of techniques to carry out weather forecasting, some more accurate than others, to predict environmental variables. Critical information about the weather is provided through weather forecasts. Accurate weather forecasting would assist the government in planning to prepare its citizens for the worst, since severe weather annually damages life, property, and many government projects that are typically substantially funded. In this study, a comparative analysis is carried out on three classification algorithms: random forest, decision tree and extra tree classifiers on weather data. Two sets of predictions were performed on weather data: the precipitation type and weather summary. With the prediction of precipitation type, the accuracy scores for both random forest and decision tree were 99.71%, while extra tree classifiers achieved 99.44% accuracy. The confusion matrix shows better prediction of records by random forest and decision tree classifiers, with a score of 99.71% each, while the extra tree classifier performed the least, predicting 99.69% records correctly. Also, the Classification report revealed that random forest had a precision of 99.84%, recall of 98.69% and F1-score of 99.26%. The decision tree classifier had a precision of 99.84%, recall of 98.69% and F1-score of 99.26%, while extra tree classifier had a precision of 99.18%, recall of 99.24% and F1-score of 99.21%. With the second set of predictions using weather summary, random forest had an accuracy score of 66.05%, decision tree had 53.9%, and the extra tree had 64.8% accuracy scores. As for the confusion matrix, random forest performed best with 66.17% records correctly predicted, extra tree classifier with 64.8% correct records and decision tree with 53.6% records. The Classification report showed random forest had a precision of 69.6%, recall of 47.8% and F1-score of 52.2%; decision tree achieved a precision of 17.39%, recall of 39.13% and F1-score of 30.43% while extra tree had a precision of 52.17%, recall of 30.43% and F1-score of 21.74%. In conclusion, the random forest classifier has proven to consistently outperform the other classifiers in this study. Further studies could focus on predicting future weather occurrences.