Page 143 - Kaleidoscope Academic Conference Proceedings 2024
P. 143

Innovation and Digital Transformation for a Sustainable World











            Accuracy










                                                                Figure 7 - Epoch vs. accuracy graph of video model
             Figure 6 - Epoch vs. accuracy graph of voice model
                                                                 Table 2 – Comparison of mean average precision
           In addition to these performance metrics, the latency of the
           voice model was evaluated to ensure its practicality in real-  Sl. No   Method   Mean Average precision
           time applications. The average latency for processing voice
                                                                 1        Tan [15]             65.38%
           inputs  and  generating  predictions  was  measured  at  1220
           milliseconds across all classes, using Python's time package   2   Jain [7]         70.17%
           for  precise  computation.  This  latency  demonstrates  the   3   Hao [16]         82.57%
           system's  ability  to  perform  timely  analysis  and  response,   4   Li [17]      82.37%
           which is crucial for real-time monitoring and intervention.    5   Yinlong [6]      85.89%
                                                                 6          Ours               88.00%
           5.2   Performance of video-based model
                                                              The implementation of the video module in proposed system
           In  the  evaluation  of  the  video-based  model,  several  key   is  targeted  to  detect  and  accurately  interpret  subtle  visual
           metrics were analyzed to gain comprehensive insights into   cues (such as facial expressions, body language, and posture)
           its  performance  and  efficacy.  Among  these  metrics,  the   at an early stage, before any further damage occurs or the
           validation loss held paramount importance as it quantified   health condition deteriorates. The impressive combination of
           the degree of error between the model's predicted labels and   high  accuracy  and  low  latency  underscores  the  proposed
           the actual labels for the validation dataset across each epoch   system's  practicality  and  effectiveness  in  real-world
           of training. A lower validation loss indicated a higher level   environment, particularly in enhancing elderly care through
           of  accuracy  and  precision  in  the  model's  predictions,   prompt and reliable voice / video based anomaly detection
           signifying an improved performance trajectory. Validation   and support. The Raspberry Pi-based system implementation
           accuracy is  used as a key  metric in assessing the  model's   facilitates versatile cost-effective healthcare solutions, which
           effectiveness  during  the  training  phase.  It  measured  the   are easy to install and maintain in a home environment.
           proportion  of  correctly  classified  instances  within  the
           validation  dataset,  serving  as  a  reliable  indicator  of  the   6.  CONCLUSION
           overall  robustness  and  reliability  of  the  machine  learning
           model. The model's achievement of a validation accuracy of   This paper presented an innovative approach to addressing
           87.91% as shown in Figure 7, underscored its capability to   the  multifaceted  challenges  of  elderly  care  through  the
           make accurate predictions across diverse data samples.    design and implementation of an elderly wellness companion
                                                              system.  The  voice  module  demonstrated  exceptional
           Precision score computed on video model, provided valuable   performance  in  analyzing  speech  patterns  and  detecting
           insights into the accuracy of positive predictions generated   emotional cues, achieving an impressive validation accuracy
           by  the  model.  Weighted  average  precision  and  macro   of 96.34%. Through meticulous evaluation of key metrics
           average  precision  were  analyzed  to  gauge  the  model's   such  as  validation  loss  and  precision,  the  voice  model
           performance  across  all  classes  in  a  typical  multi-class   exhibited  robustness  and  reliability  in  discerning  and
           classification  scenario.  With  both  metrics  yielding  a   classifying different speech patterns and emotions. The voice
           consistent precision score of 0.88, the model demonstrated a   module acts as a preliminary check, which triggers the video
           creditable level of accuracy and consistency across diverse   module  only  when  severity  in  state  is  detected,  ensuring
           classes, affirming its suitability for real-world applications in   camera privacy of the elderly. Furthermore, the video model
           healthcare and beyond. The comparative analysis of Mean   complements the voice-based anomaly detection system by
           Average  Precision  with  existing  systems  is  tabulated  in   providing additional insights into the physical state of elderly
           Table 2.                                           individuals. Through the seamless integration of voice and
                                                              video-based anomaly detection technologies, the proposed
                                                              system  offers  comprehensive  and  holistic  care  for  elderly
                                                              individuals, promoting independence, safety, and well-being.





                                                           – 99 –
   138   139   140   141   142   143   144   145   146   147   148